Blog

What was said

At AMLA’s first conference on 9 June 2026, the Commissioner for Financial Services, Maria Luís Albuquerque, made an argument about the form of the rules rather than their content. When Member States implemented the old framework differently, she said, the Union’s answer was to make the rules more detailed and leave less room for manoeuvre. That, in her account, added layers of complexity without adding effectiveness. A consistent European approach to supervision through AMLA now makes it possible to move the other way, towards more principle-based rulemaking and risk-based supervision.

She anticipated the objection. Many stakeholders hear simplification and fear deregulation, she said. But complexity, she said, does not necessarily produce better supervision, and a simpler framework is in many cases a more effective one. Good anti-money laundering supervision, in her words, “rests on effective risk assessment and sound judgement”, and “is not measured nor necessarily achieved by a (high) volume of reporting”.

The German State Secretary for Finance, Jeanette Schwamberger, spoke the same morning for the host Member State. She set four conditions for implementation. Requirements should be met as simply as possible. Guidance should be precise and practical. The risk-based approach should be applied consistently. And small and medium-sized firms in the non-financial sector need particular care, because they do not have the compliance departments that banks do.

AMLA’s own Chair had set the boundary in her opening address. The Authority will examine every instrument for opportunities to simplify, and will give a reasoned justification of how proportionality has been applied. But simpler and more proportionate must never come at the expense of effectiveness. She did not leave it as a caveat: anti-money laundering and counter-terrorist financing, she said, are too important for compromise, and the bar must remain high.

Closing the same conference, the Vice-Chair, Juan Manuel Vega Serrano, drew the line the Commissioner had drawn, from convergence to simplification. Convergence is being built now, he said, while the remaining standards and guidelines are written, and simplification will be embedded in that work. So the Authority itself connects the two, with its Chair’s condition attached.

What it is not

It is not a measure. Nothing had been proposed at the time of writing, and the Commission’s work programme for 2027 contains no anti-money laundering initiative at all. The Regulation applies on 10 July 2027 as written. The standards AMLA has already sent to the Commission are detailed instruments, and nothing said in June changes them.

So no firm should plan on lighter obligations arriving before the Regulation applies. What June did was put a direction on the record, from the institution that drafts the legislation and from the largest Member State, without either needing the other to say it.

Where it could arrive

Three routes exist already, and none requires anything new to be invented.

AMLA’s own review mandate. Since Regulation (EU) 2025/2088 amended the AMLA Regulation, the Authority has had a standing task on simplification. It assists the Commission in reviewing how the technical standards are applied, and it reviews its own guidelines. It also proposes amendments that remove redundant or obsolete reporting requirements and make reporting proportionate. That is already in force. A political push for simplification has a channel waiting for it.

The banking package. The work programme announces a bank competitiveness and simplification package for the first quarter of 2027. Its legal basis is Article 114 TFEU, the same one the anti-money laundering package was built on, and it is addressed to the same institutions. Nothing says it touches anti-money laundering. It is the kind of vehicle that could, and it arrives some months before the Regulation applies.

The Pact against gold-plating. Also in the work programme: a commitment that Member States be transparent about what they add when transposing Union rules, with a review and remove mechanism in the European Semester. The sixth Directive must be transposed by 10 July 2027. Gold-plating at transposition is exactly where the same obligation starts to cost differently from one Member State to the next.

The experiment that has already been run

This framework has one worked example of what principle-based rulemaking does in practice, and it is recent. AMLA consulted on its standards on group-wide requirements in the spring of 2026 and published a final text that was shorter and far less prescriptive than the draft. A waterfall test for which entity carries the group obligation became a set of factors to be weighed. A demand to assess and consolidate every entity’s risk profile became a holistic group-wide view. Five prescribed third-country scenarios became one set of minimum actions plus tailored measures.

The industry had asked for exactly that and largely got it. As an earlier post here put it, a formula needs a sum and a judgement needs a file. Under the waterfall nobody had to justify the outcome, because the test produced it. Under the factors, the notification to the supervisor has to carry detailed reasons on each of three tests. The work did not shrink. It moved from the rulebook into the firm’s own records.

That is the likeliest shape of whatever simplification follows. Fewer pages of rule, more pages of reasoning, and the reasoning held by the firm rather than printed in the Official Journal.

Who carries the judgement

The Commissioner’s own words locate it. Good anti-money laundering supervision, she said, rests on effective risk assessment and sound judgement. Sound judgement is exercised by someone. In a principle-based regime it is exercised twice: once by the firm deciding how a principle applies to its business, and again by the supervisor deciding whether the firm decided well.

That second judgement is where the risk sits for a firm. A detailed rule tells you in advance what the supervisor will check. A principle tells you what the supervisor will ask, and the answer is whatever your file can show. The equivalent safeguard for a firm under principle-based rules is a record that shows how it reached each judgement and why.

None of this is an argument against simplification. A rulebook that grows each time national practice diverges has costs of its own. The Commissioner’s point was that complexity does not necessarily produce better supervision. It is an argument about who does the work once the detail is removed.

What a firm can do now

Plan for the Regulation and the standards as written. Nothing announced at the time of writing changes them, and the date does not move.

Then look at the decisions your current processes take by following a rule rather than by reaching a view. Examples are customer risk ratings produced by a matrix, thresholds applied because the text names them, and group allocations settled by a test. If those were to become judgements, could you show how each was reached? That is the capability simplification would demand, and it can be built now at no cost to compliance with the rules as they stand.

And respond to the consultations. The Commissioner asked the room to do exactly that, and in a regime moving towards principles, consultation is where the content of the principle is still being settled. After it, the content is applied through supervisory practice.

What has been published

AMLA’s webinar slides, issued in corrected form on 16 June 2026, are among the most operational documents the Authority has published on selection. They say repeatedly that AMLA will supervise up to 40 institutions, and that up to 40 high-risk eligible entities and groups will be selected. AMLA staff repeated the figure when presenting them on 10 June 2026. The slides also set out a quarter-by-quarter plan that appears nowhere in the Regulation. Provisional eligibility is assessed in the third quarter of 2026. Risk assessment and eligibility data collection follow in the first quarter of 2027, risk assessment in the third quarter of 2027, and selection in the fourth. Direct supervision begins in the second or third quarter of 2028.

For the first round, the figure is right. What the slides do not say is that it is a rule for one round.

Two rules, not one

Article 13 sets the standing rule, and its first two paragraphs are short enough to read in full.

Article 13(1): Credit institutions and financial institutions, and groups of credit institutions and financial institutions, whose residual risk profile has been classified as high pursuant to Article 12 shall qualify as selected obliged entities.

Article 13(2), first subparagraph: However, where more than 40 entities are identified pursuant to paragraph 1, the Authority may, in consultation with the supervisory authorities, agree on limiting the selection to a specific different number of entities or groups that is greater than 40.

Read alone, that makes forty a floor. Qualification involves no discretion: an entity classified as high residual risk shall qualify. Forty appears only as the point above which the Authority may limit the list, and any limit it agrees must be greater than 40.

But Article 13 does not stand alone for the first round. Among the transitional arrangements, Article 106(2) provides: By way of derogation from Article 13(2), where, during the first selection process, more than 40 obliged entities would qualify for direct supervision pursuant to Article 13(1), the Authority shall carry out the tasks listed in Article 5(2) in respect of the 40 obliged entities or groups operating in the highest number of Member States whether through establishments or under the freedom to provide services.

So for the first list forty is a ceiling, set by the Regulation rather than chosen by the Authority. The recitals give the reason: the Authority should have the resources to supervise up to forty groups and entities at the same time, at least during the first selection process. If forty or fewer are classified high, all of them qualify. If more are, the forty operating in the most Member States are taken, and among those operating in the fewest, the ratio of third-country transactions to total transactions breaks the tie.

From the second selection the derogation no longer applies and Article 13(2) governs. The first selection must begin by 1 July 2027 and the process repeats every three years from that date, so the second begins in 2030.

The panel and the second round

Executive Board member Derville Rowland moderated the panel on supervision and cooperation at AMLA’s first conference on 9 June 2026. On that panel, Executive Board member Rikke-Louise Ørum Petersen said that the forty entities are the foundation, that there will be several rounds, and that more may be added. On the text, from the second round onwards any cap must be greater than forty.

Read with Articles 13(2) and 106(2), the two statements need not conflict: the slide fits the first round, and the panel remark fits the rounds after it. The gap is in how the slide has been read: as a permanent size for direct supervision rather than the size of the first list.

Ørum Petersen said one other thing. Appearing on the list reflects systemic importance and cross-border reach, and, in her words, is not a sign of weakness.

The first cap rewards a narrow footprint

Look at how Article 106(2) chooses, if more than forty qualify. It does not take the forty with the highest risk. It takes the forty operating in the most Member States. A firm classified high and operating in seven Member States can be left out of the first list while a firm operating in twenty is taken.

That firm has not been cleared. Its classification stands, and the same methodology is the one its national supervisor works from. In 2030 the cap, if the Authority sets one at all, must be greater than forty, so the same classification is more likely to be enough. A firm that misses the first list on the count of Member States has been deferred, not excluded.

Every Member State gets a second look, from 2030

Article 13(3) has had the least attention of any part of this mechanism. A Member State may have no credit institution, financial institution or group that qualifies under the first two paragraphs, counting those established, authorised or registered there and those with a subsidiary there. In that case the Authority shall carry out an additional selection process in that Member State. It does so on the basis of the methodology in Article 12(7), point (b), which classifies risk profiles as low, medium, substantial or high. Following it, the entity there whose risk profile is classified as high qualifies. If several are, the one operating in the most Member States qualifies, with the third-country transaction ratio as the tie-break.

Two limits on that. It is not a guarantee of one entity per Member State: the additional process still ends in a classification, and only a high one qualifies. And Article 106(3) disapplies it during the first selection process, so it operates from the second round.

From 2030, its effect is that a firm classified high in a Member State with no selected entity is not ranked only against the most cross-border institutions in the Union. It is looked at again within its own market. The paragraph does not restate the six Member State condition in Article 12(1), and nothing AMLA has published says how it reads that.

Who is in the population being assessed

Two features of Article 12 widen it beyond a count of offices.

The periodic assessment covers entities operating in at least six Member States, including the home Member State. Activity counts whether it runs through establishments or under the freedom to provide services, and regardless of whether the activities are carried out through infrastructure on the territory concerned or remotely. Remote provision counts. A firm with one office and customers in twelve Member States may be inside the assessment. It will be if its activity in at least six of them clears the materiality thresholds in AMLA’s draft standard. And counting to six is wider than it looks.

And the methodology is to be established separately for at least eleven categories of obliged entity. They are credit institutions, bureaux de change, collective investment undertakings, credit providers other than credit institutions, e-money institutions, investment firms, payment institutions, life insurance undertakings, life insurance intermediaries, crypto-asset service providers, and other financial institutions. A separate methodology for a category is not something the Regulation would require for a population nobody expects to assess.

AMLA has added a practical limit for the first round. Its webinar materials of June 2026 tie the 2027 process to the eligibility data collected in 2026. They also require every entity that receives the template to return it, even one that is plainly not eligible. The Regulation defines the population by the six Member State test, and Article 12(2) requires entities subject to the assessment to supply the information it needs. A firm that meets the test and has not been asked for anything has reason to raise it with its national supervisor rather than wait for 2030.

Getting out is harder than getting in

Under Article 13(6) a selected entity remains under direct supervision until the Authority commences supervision under a subsequent list that no longer includes it. Not when the new list is published, and not at the end of a cycle. The exit comes when supervision under the next list actually starts, which under Article 13(4) is six months after that list appears.

So a firm that drops out of the second list stays directly supervised for the whole of the gap. The exit comes six months after the next list is published, not on publication. Because entry is lagged by the same six months, the period of direct supervision is still about three years.

What this does not mean

It does not mean the second list will be large. Article 13(2) requires the Authority to take its own capacity into account when setting a number, and to ensure the increase in staff and budget is feasible. A second list only modestly above forty is therefore entirely plausible. Nor does it mean the first list will reach forty: if fewer are classified high, it will be shorter. Nothing here predicts a figure.

It also does not mean being outside the list is a reprieve. That argument was made at length when the list first drew attention and it has only strengthened since. AMLA’s calibration exercise uses a common sample for the selection methodology and for the methodology national supervisors will use, so the two are being built to stay consistent. The list decides who holds the file. It does not decide what the file is judged against.

What a firm can do with this

Use the number for what it decides, which is the first list. If the conclusion that direct supervision is somebody else’s problem rests on ‘up to forty’, it holds for 2027 at most. It should be retaken for 2030 on the criteria that govern then. There are two: operating in at least six Member States, counted the wide way, and a residual risk classification produced by a methodology that is still being calibrated.

Then look at the data return, because that is the part a firm controls. Eligibility information was collected as at 31 December 2025 and submitted by national supervisors to AMLA by 15 August 2026, and risk assessment data collection falls in the first quarter of 2027. At the June conference a representative of a large cross-border banking group described what that looked like from the inside. Delivery was decentralised to national level, so the group required its own entities to submit to the centre first and compared them against each other before anything was returned. The comparison found its own subsidiaries applying different definitions of a financial institution despite a group standard.

That is the practical lesson and it is available to any group facing the same template. The return goes back country by country, so nothing reconciles it unless the group reconciles it. A classification produced from inconsistent inputs may not look wrong from outside. It is simply the number the model receives.

What is not established

How many entities are expected to be classified high in the first round is not published, and that number decides whether the Article 106(2) cap bites at all. The risk assessment methodology is still in calibration, so which entities fall into the high residual risk class cannot be anticipated from outside. Whether the Authority will cap the second list, and at what number above forty, is a decision for 2030.

What the work programme contains

The Commission work programme for 2027 is COM(2026) 710 final, adopted on 4 November 2026, with four annexes: new initiatives, priority pending proposals, withdrawals and envisaged repeals.

The Communication does not mention money laundering, financial crime, corruption, anti-corruption or sanctions. Not a passing reference, not a sentence of context. And none of the twenty-three initiatives in Annex I is an anti-money laundering measure.

Why that is not neglect

The legislative work is finished. The Regulation and the sixth Directive were adopted in 2024. What is left for 2027 is Level 2: regulatory and implementing technical standards, guidelines, delegated acts. A work programme lists the flagship initiatives a Commission intends to propose. It does not list the standards that put them into effect, because those are not proposals in that sense.

So the silence looks structural rather than editorial. There is no sign that anyone decided to leave it out.

It is also not new. The work programme for 2026, COM(2025) 870 final of 21 October 2025, does not mention money laundering either. Two programmes, two years, covering the run-up to the framework and the year it starts to apply, and neither mentions it.

Which is precisely the thing worth saying. The document that tells the public what the Union will do next year is silent about the largest change to financial crime compliance in a decade, in the year it takes effect. That is because the remaining decisions have moved into instruments the document does not cover. The fifty-four instruments in the tracker are, to the one document most people would check, invisible.

What is in it that reaches you anyway

A banking package in Q1 2027, and it is bigger than its name. The work programme announces a Bank competitiveness and simplification package, legislative, with an impact assessment, under Article 114 TFEU. Annex III then withdraws the 2015 proposal for a European Deposit Insurance Scheme on the stated ground that it is to be replaced by the proposals on banking competitiveness in Q1 2027. So that package carries the deposit insurance file. It arrives in the quarter before the Regulation applies. It rests on the same legal basis as the anti-money laundering package, is aimed at the same institutions, and sits under a heading of simplification and reduced administrative burden. Nothing says it touches anti-money laundering requirements. It is the kind of vehicle that could, and it is worth reading when it lands rather than after.

One third of the payments package is listed for withdrawal. The June 2023 package had three parts. Two are at second reading. The third, the framework for Financial Data Access, is listed for withdrawal within six months: pending more than three years, no trilogues since June 2025, unlikely to secure Council support. If an open-finance plan in your firm assumed all three parts would arrive, one of them is now listed for withdrawal.

The anti-fraud architecture review finally has named outputs. The Communication says the protection of the EU budget will be enhanced by revision of the OLAF and EPPO Regulations and the PIF Directive, as a result of that review. Three instruments rather than one, and none of them scheduled: no quarter, no entry in Annex I. A statement that they are coming, not a date on which they come.

And the digital euro is a priority. It sits in Annex II with its two companion proposals, and the Communication promises support for a swift agreement. That is intent rather than a date, and it says nothing about the offline cap, which press reports describe as the point where the negotiation has stalled.

The one sentence with a direct bearing

Buried in the simplification section: to avoid rules agreed at Union level becoming more complex when transposed by Member States, the Commission will work towards a Pact against gold-plating. Member States would be transparent about what they add, and a review and remove mechanism would sit in the European Semester.

The sixth Anti-Money Laundering Directive must be transposed by 10 July 2027. Transposition is where twenty-seven national frameworks diverge, and divergence is what makes the same obligation cost differently in Dublin and in Milan. A Pact against gold-plating is aimed squarely at that, in the same year. Whether it amounts to anything is another matter, and nothing in the document says it will reach anti-money laundering transposition in particular.

What to take from it

Stop treating the work programme as an early warning system for this framework. It is a good document and it is the wrong document. It will not carry the instruments that decide your obligations in 2027, or in 2028.

The early warning now sits in three places: AMLA’s consultations, the General Board summaries, and the Commission’s delegated and implementing acts as they appear. None of them is published on a schedule anyone announces. That is a harder watch than reading one Communication in November.

What the Regulation requires

The obligation sits in the Regulation that created AMLA. Article 84(4) of Regulation (EU) 2024/1620, under the heading Accountability and reporting, says this:

Within six weeks of each meeting of the General Board, the Authority shall at least provide the European Parliament with a comprehensive and meaningful record of the proceedings of that meeting that enables an understanding of the discussions held within that meeting, including an annotated list of decisions. Such record shall not reflect discussions within the General Board relating to individual obliged entities or discussions relating to confidential supervisory or FIU-related data, unless otherwise provided for in the legislative acts referred to in Article 1(2).

The General Board’s own Rules of Procedure reproduce that paragraph at Article 6(19) almost word for word. The Rules are a Decision of the Board, done at Frankfurt on 6 March 2025 and signed by the Chair. One thing is added in the copying: the Rules provide that the record is submitted for approval to the members of the Board, which the Regulation does not require. That addition need not be read as anything more than a board restating in its own procedural terms a duty imposed on it from above. The version that binds is the Regulation’s.

Two small words carry weight. “At least” sets a floor rather than a ceiling. And the record Parliament gets is itself redacted, so this is not a complete version against a censored one. There are three layers: what was said, what Parliament is to receive, and what is published.

What comes out

Article 6(20) of the Rules governs the public document. After the record has gone to Parliament, a summary is published on AMLA’s website, without the annotated list of decisions and without the excluded material. Article 8(3) makes all other proceedings of the Board confidential.

And this is the part worth sitting with. Nothing in the Regulation requires that summary to exist. Article 84 sets out what AMLA owes. There is an annual report, made public and presented in public to Parliament, and a hearing before the competent committees at least once a year. Parliamentary questions must be answered within five weeks. Confidential discussions are held behind closed doors on request, and the record of proceedings goes to Parliament within six weeks. Article 60(7) adds that the General Board must adopt its rules of procedure and make them public. A public account of what the General Board discussed is in none of it. It exists because the Board decided it should.

So what the public version leaves out is the annotated list of decisions, the part a reader would most want. The summaries still report conclusions, but not as a formal list. What remains is the discussion, in prose, as summarised by the Authority.

This may explain a difference in wording between the two provisions. Article 6(19) says the Authority shall provide the record to Parliament. Article 6(20) says a summary should be published. The first is a duty imposed by the Union legislature and copied down. The second is the Board writing about itself, and Article 6(20) says the summary “should” be published.

It is visible in the annexes

The mechanism surfaces where housekeeping is listed. Among the written procedures annexed to the fourth meetings in October 2025 are approvals of records of proceedings for the European Parliament, from the meetings of May and June 2025. The Board approves the Parliament version by written procedure, and the public learns of it from a line in a list.

That also marks the limit of what can be said. For those two meetings the step is documented. For the other meetings, if this page reads the annexes correctly, the step does not appear in them. Is it simply recorded elsewhere? Very possibly: the annexes are lists of written procedures, not a register of what was sent to Parliament.

What this is not

It is not a complaint about secrecy, which would be cheap and also wrong. The exclusions protect named firms and financial intelligence data, and a reader who thought about it for a moment would want them there. A supervisory authority that discussed an individual bank in a published document would be doing something worse than withholding one.

The point is narrower. The public document is a redaction by design rather than by accident, the redaction is of the decisions specifically, and a fuller version exists and sits with the committees that scrutinise AMLA. Anyone relying on these summaries is working from the second-best version.

What follows from it

A voluntary disclosure can be narrowed without anyone breaching anything, and the format of this one has changed. The summary of the first meeting, in March 2025, named every representative of every Member State in both compositions, fifty-four entries covering fifty-two officials, and marked who attended online. The recent ones attribute every position to Members collectively and record no vote, no name and no formal list of decisions. Does the change in format reflect a deliberate policy, or simply a settled template? Nothing published says, and nothing in the Regulation would prevent either.

So read them closely, and do not mistake the public version for the record. What binds is the duty to Parliament. What everybody actually reads is a voluntary publication.

What AMLA set out in the Board

At the General Board in joint composition on 30 June and 1 July 2026, the Chair introduced the item. An Executive Board member then set its limits. The purpose was to agree a direction rather than a text, the subject is close to fundamental rights, and AMLA cannot authoritatively interpret Level 1 legislation.

The analysis that followed had three parts. Legal privilege is protected under Union law and the European Convention on Human Rights, subject to restrictions that are necessary and proportionate. It applies to the specific categories of information defined in the Anti-Money Laundering Regulation. And, in AMLA’s understanding, it should not be treated as a general exemption from anti-money laundering obligations, customer due diligence, the reporting of suspicions and information-sharing requirements included.

The second of those three is the operative one, and it is easy to read past. Privilege attaching to categories of information defined in an instrument is a different thing from privilege attaching to a professional relationship. On the first reading, the question is always what this document is. On the second, the question is who you are.

Who disagreed, and on what ground

A number of Members expressed significant reservations. Their argument was that privilege derives primarily from fundamental rights protected by the Charter and from longstanding national legal traditions, so its scope may be broader than AMLA allows. They urged caution, questioned whether any public statement was appropriate or timely, and said further technical discussion should happen in AMLA’s committees with the legal profession and self-regulatory bodies present. Some proposed a questions and answers format rather than a formal statement.

Other Members supported AMLA, on the ground that privilege read as a blanket exemption would undermine reporting, supervision and the work of the financial intelligence units.

So the recorded disagreement is among the members of the Board. Some of them asked for the legal profession and self-regulatory bodies to be present when the discussion continues.

What the Commission said, and what was decided

The Commission agreed with AMLA, taking the view that the relevant provisions largely continue existing Union legislation and create no general derogation. That matters because the Commission adopts the regulatory technical standards, although any guidance on this point would be AMLA’s own.

And then nothing was decided. The Chair sent the question back into the committees, and the Board supported continuing the discussion in the supervisory and financial intelligence substructures before any guidance or communication is considered. The item ended without a position.

Why that makes it more useful, not less

A practitioner could reasonably say that an unresolved discussion at a board meeting changes nothing, and today that is correct. The reason to read it anyway is that it tells you where the pressure is coming from before it arrives as a document.

Two things are now on the record that were not before. AMLA has set out, in its own Board, a reading of privilege that is narrower than the reading of some of its members. And the institution that adopts the binding standards agrees with it. Neither can interpret Level 1 by fiat, which is why nothing has been published. But standards, guidelines and supervisory expectations are all written by people who hold a view, and the view is known.

One more fact to hold alongside it

At the same round of meetings, AMLA set out how it proposes self-regulatory bodies take part in its internal committees, and the Board took note. The proposal is ad hoc participation, dedicated consultation formats and structured information sharing, within the confidentiality rules. Consulted, in other words, and not seated.

In many Member States those bodies are the supervisor for the professions and the source of the guidance practitioners actually follow. On the question of how far privilege reaches, they will be asked rather than present.

What a practice can do now

Nothing here is a reason to change a position, and this is not advice to change one. It is a reason to know the size of the thing that would be affected if the narrower reading prevails.

Start with a count. How many current matters are being handled on the basis that the anti-money laundering obligations do not apply at all, as distinct from applying with a carve-out on reporting? Many practices may not be able to answer that quickly, because the decision is often a practice-wide assumption rather than a recorded, per-matter judgement. That population is the exposure.

Then look at what the claim rests on, file by file. If the basis recorded is the status of the relationship, it is resting on the broader reading. If it identifies which category of information under the Regulation the claim attaches to, it survives either reading. The difference costs nothing to record now and is very expensive to reconstruct later.

And settle who decides. In some practices the privilege call is made by the partner holding the matter, which means it is made differently in different corners of the same firm. One named person, one standard, one note on the file. If the direction on the record becomes a document, the firms that will have least to do are the ones that can already show how each call was made.

The tracker entry for this shows the position: no instrument, no draft and no published position yet.

What AMLA has committed to

At the General Board in supervisory composition on 1 July 2026, AMLA set out how it intends to discharge its sanctions-related mandates. Its starting point is that the new framework builds the mitigation of risks from non-implementation and evasion of targeted financial sanctions into three core processes: risk assessments, customer due diligence and ongoing monitoring.

Note where those risks are being put. Not into a sanctions annexe. Into the processes the anti-money laundering function already owns.

The conclusions were unusually specific for a Board summary. A dedicated task force under both internal committees. High-level expectations on sanctions written into upcoming instruments, as has already been done in the drafts prepared so far. A comprehensive set of guidelines on targeted financial sanctions by the second quarter of 2027. Coordination with the European Supervisory Authorities, the EBA in particular, and with the European Central Bank. And an investigation of how to bring external national authorities, ministries of finance among them, into the task force without breaching confidentiality.

The date, read properly

The second quarter of 2027 runs from April to the end of June. The Regulation applies on 10 July 2027. So the comprehensive guidelines on sanctions arrive somewhere between about three months and ten days before the framework they relate to takes effect.

A firm that waits for them will be reading them in the same weeks it is going live. That is not a reason to criticise the timetable, which is tighter for AMLA than for anyone. It is a reason not to plan around it.

The expectations are already arriving

The guidelines are late in the sequence, but the expectations are not waiting for them. Targeted financial sanctions already appear in the customer due diligence standard, in the group-wide standard and in the business-wide risk assessment guidelines. AMLA said as much: high-level expectations have already been written into the drafts prepared so far.

So the sequence is the wrong way round from a firm’s point of view. The obligations are landing in instruments that are already final or nearly final. The document that explains how to meet them comes last.

What the Board was not sure about

The direction is not unanimous, and the disagreement is useful to know. Some members questioned whether a standalone set of guidelines is needed at all, rather than folding the elements into broader instruments. Several asked that the approach stay principles-based, given how fast the sanctions environment moves. Others questioned what AMLA’s supervisory role here will actually be, and whether national sanctions authorities should be involved at all.

Those are not procedural quibbles. Sanctions designation and enforcement are not AMLA’s responsibility. AMLA’s own strategy keeps the Union’s restrictive measures regime distinct from the anti-money laundering supervisory framework. What AMLA is claiming is narrower and still significant: the risk of a customer being used to evade sanctions is an anti-money laundering risk, and it will be supervised as one.

In the joint composition the same work appears from the other side. It runs through a Task Force on Cooperation, with Member State representatives and observers from the EBA and the European Central Bank. It also includes guidelines on the evasion of targeted financial sanctions. As far as this page can see, these would be the first AMLA text addressed to obliged entities on sanctions risk. That pair carries no date at all, which is a second reason not to treat the second quarter of 2027 as the moment everything becomes clear.

Why this lands on the organisation chart

Evasion is the awkward word. A customer structured to move value for a designated person is a sanctions matter to the sanctions team and a money laundering typology to the anti-money laundering team. In some firms it is nobody’s in particular, because each assumes the other owns it. Guidance that builds evasion risk into risk assessment, due diligence and ongoing monitoring gives that ambiguity somewhere to land, and the place it lands is the anti-money laundering function.

That function cannot discharge it from a separate system. Screening results, designation hits and near-misses sit in one estate; customer risk ratings, due diligence files and monitoring alerts sit in another. Neither is wrong. They were built to answer different questions. The guidance is likely to assume a firm can answer both about the same customer, at the same time, from a record that shows when each was last reviewed.

What to do in the months you have

Four questions, in order of how much they cost to answer.

For a single named customer, can the firm produce the sanctions screening history and the anti-money laundering risk rating together, each dated, without someone exporting from two systems and reconciling by hand? If the answer requires a person, it will not survive a supervisory request at scale.

Does a sanctions near-match change the customer’s anti-money laundering risk rating, and does an anti-money laundering suspicion change how that customer is screened? In many firms the answer to both is no, and there is no technical reason for that beyond the two estates never having been joined.

Who decides when the two functions disagree about the same customer, and is that person named anywhere? Escalation routes that run in parallel to two different committees produce two defensible decisions and no single answer.

And the one that matters most: who owns evasion risk today. If the honest answer is that it is shared, it is unowned, and the firm has until the second quarter of 2027 to give it a name.

Why the Board can change a draft at all

AMLA explained the mechanics itself, at the meeting of 5 February 2026. Consultation items come to the Board as points for discussion rather than decision. This is done to confirm the Board’s support for launching a consultation and to keep the drafting moving in parallel. The Board’s final decision-making power, exercised by qualified majority, is left intact. Comments made in the meeting are captured and reflected in later versions and in the process leading to adoption.

So an agenda marked for discussion is not a briefing. It is the stage at which the people who will later vote say what they want, to the people still writing.

One: a word was added

Following agreement by the Board on 5 February 2026, the enforcement standard was amended so that failures in anti-money laundering systems, controls or policies must be material. The same clarification was made at Article 5(3)(d). That standard is now in final draft, awaiting adoption by the Commission.

One adjective, inserted at a Board meeting, between a firm and a finding of breach. It is the clearest instance in these summaries of the Board changing a text rather than noting it.

Two: a provision was contested and kept

At the same meeting, three Board members objected to the inclusion of Article 17 of the customer due diligence standard and to its potential effect on the funds industry. Others defended it as a balanced compromise that avoids systematic look-through to the investors in a collective investment undertaking and preserves the fund manager’s ultimate responsibility.

Three Member States wanted it out and it survived. What changed instead was the surrounding language: newly inserted recital wording saying the frequency and extent of any request should reflect the specific risk being addressed. If you administer funds, that recital is the sentence to keep, and it sits beside the provision three supervisors objected to.

Three: a consultation was opposed and launched anyway

On 1 July 2026, on the baseline risk assessment model for the non-financial sector, members argued for more proportionality through simplified methodologies, population-based assessment, sampling and longer transitional arrangements. The timeline was questioned. Two members opposed proceeding and a third recorded reservations.

The Chair proceeded. The timeline was to be revisited and specific questions were added on alternative approaches that could achieve the same objectives proportionately. Opposition did not stop the consultation; it changed the questions the consultation asks. That is a smaller win than blocking, and it is not nothing: the questions decide what evidence comes back.

Four: an ambition was refused

At the same meeting AMLA asked whether it should develop common principles for the final step in setting a sanction. Its argument was that sanctioning practices and sanction levels differ significantly between Member States. That, it said, produces inconsistent outcomes and creates opportunities for regulatory arbitrage.

Most members did not want it, at least not yet: implement the standard and the base-amount guidelines first and see whether they work. The scope was not extended. The Authority named a problem in its own supervisory architecture and asked for the power to address it. Most of the national authorities told it to wait, because they wanted the existing instruments tested first.

Five: a direction was set on an item listed for information

This is the one to read twice. On 29 April 2026, the Article 69(3) reporting standard was on the agenda for information only. Even so, the Board pressed it towards a minimum mandatory core with optional fields and a longer roadmap. AMLA staff undertook to reflect that preference in the drafting.

The consultation launched in July carried exactly that shape: common templates, a phased implementation approach, a minimum mandatory core. The design of the instrument firms will report suspicions through was settled at a meeting where the item was not even down for decision, two months before anybody outside saw a draft.

What the five have in common

One word, one provision, one timetable, one refusal and one design. None of them is recorded as arising from a consultation response. In four of the five the consultation had not opened yet, and in the fifth the change was made to the text that went out.

All five are in documents AMLA publishes. They are not leaks and they are not inference. They are in the summaries, which run to a few pages each and which few in the industry read.

What to do with this

Read your own supervisor’s positions. The summaries do not name Member States, so you cannot always tell who said what, but you can tell what was argued, and you can ask. A supervisory relationship includes the ability to say: the Board discussed proportionality for our sector in July; what position did you take?

And move your argument earlier. In the fifth instance, the shape of the draft had been agreed before the consultation opened. The consultation is the right place to argue about wording. It is the wrong place to argue about design, because the design was settled in a room your supervisor was sitting in, about two months before.

Where the gap was identified

At the General Board in supervisory composition on 1 July 2026, several members raised the handling of ongoing supervisory and enforcement procedures during the transfer of competence. Their concern was specific. National supervisors may be unable to continue proceedings once competence passes to AMLA. Some administrative procedures run for a long time, so the absence of a clear legal mechanism creates legal uncertainty and operational difficulty.

That last clause is the part a lawyer will notice. The members raising it were not describing a theoretical boundary problem. The summary does not say whether they had particular procedures in mind.

Why it cannot be fixed where everything else is being fixed

AMLA’s answer was that the Commission had taken the view that Level 2 measures cannot alter the allocation of competences established at Level 1. Alternatives had been explored during drafting and none was compatible with that constraint.

The reasoning is sound and the consequence is hard. Almost everything else firms are waiting for in this framework is being settled in technical standards and guidelines, which is why so much of the detail is still moving. This cannot be. Who holds supervisory competence over a given entity is decided in the Regulation itself, and a standard made under the Regulation cannot move it. Fixing this properly would mean amending Level 1, and nothing of the kind has been proposed.

What is left is what AMLA described: encourage completion during the transition, and handle whatever remains through bilateral cooperation between AMLA and the national authority. Several members accepted this as not ideal and the best available. The record says so plainly.

Read that as a firm, not as an institution

Bilateral cooperation between two authorities is a sensible answer to an institutional problem. It is a poor answer to a procedural one, because a procedure is not only a matter of who acts. It is a matter of what the subject of it can expect: what will be put to them, in what order, with what opportunity to answer, and with what route of challenge.

Two authorities coordinating informally over an open file can reach an outcome. What they cannot do is supply the certainty that a procedure normally supplies. That certainty comes from a legal mechanism, and the record says there is not one.

The second half of the problem

There is a companion gap, and the two should be read together. AMLA’s Administrative Board of Review carries out an internal administrative review of decisions AMLA takes under Articles 21, 22, 23 and 77 of its Regulation: administrative measures, pecuniary sanctions, periodic penalty payments and fees. The Regulation requires its five members and two alternates to be appointed after a public call for expressions of interest in the Official Journal. No such call appears to have been published, and the body does not appear on AMLA’s governance page.

Put the two together and the window has a shape. It begins in about eighteen months. In it, a firm may face a procedure whose owner is not settled. Unless the Board of Review has been appointed by then, the firm may also face an authority whose internal review body is not yet in place.

What a firm can do about it

Two questions, both answerable this quarter and neither requiring anything from a supervisor.

What is the longest-running administrative or enforcement procedure currently open against the firm anywhere in the Union, and when did it start? Firms tend to know their largest exposures and lose track of their oldest ones, and age is what matters here rather than size.

Would it still be open in mid-2028 on its current trajectory? If the honest answer is yes or probably, the firm is in the population this affects. The useful response is not to wait for a mechanism that, according to AMLA, the Commission considers cannot come from Level 2. It is to treat completion before transfer as something worth actively seeking, through whatever cooperation the procedure allows, rather than something that will be handled.

One further thing worth doing, and it costs nothing. Keep the firm’s own continuous record of the procedure: what was asked, when, by whom, what was supplied and what was said in response. If a file does pass between two authorities without a mechanism designed for the purpose, the version that is complete and dated may turn out to be the firm’s own.

What the standard does tell you

Two numbers are published and citable, in the final draft standard AMLA submitted to the Commission, which the Commission has not yet adopted. They decide whether activity provided across borders into a Member State counts towards the six-Member-State test. It counts if there are more than 20 000 resident customers there, or more than EUR 50 000 000 of their incoming and outgoing transactions. An office, branch or subsidiary counts without any threshold. They sit in the Article 12(7) standard, and Counting to six set out how they are counted and why the answer is rarely the one a firm expects.

Eligibility is the easy half. It puts a firm into a pool. What happens next is scoring: every entity in the pool is assessed against a methodology, each one ends up with a risk profile, and the profiles decide which entities AMLA actually takes.

What it does not tell you

On 10 December 2025, approving both risk assessment standards for submission to the Commission, the General Board in supervisory composition recorded two things in the same breath. Targeted clarifications had been made to support a timely start to the selection process. And certain elements, such as the indicator weights and the thresholds, would remain outside the standards, to allow future adjustments.

That is the finding, and it is stated plainly enough in the Board’s own summary that it needs no interpretation. The scoring exists. It is being built. It is not in the instrument.

Why outside the standard is not an administrative detail

The objection period is not a formality. It is the point at which a technical standard stops being the authority’s document and becomes everyone’s. A draft is consulted on, a final report explains what the consultation changed, the Commission adopts, and then two institutions that answer to electorates can say no. A firm that dislikes a threshold has, in principle, four places to say so.

If this page understands the position correctly, a number that sits outside the standard would have none of those places. It does not appear in the consultation, because it is not in the draft. It does not appear in the final report, because there is nothing to report. It is not adopted, so it cannot be objected to. And nothing published says whether a change to it would be announced.

If that is right, three consequences follow for a firm, and they are practical rather than constitutional. It cannot model its own score, because it does not have the weights. It cannot tell whether the test has moved since it last looked, because a movement leaves no trace. And it may have no procedure in which to argue.

This is not theoretical

The numbers were still being set nearly seven months after the standards were approved. On 1 July 2026 the Board was told that national authorities would take part in the discussions on risk indicators, thresholds and methodological adjustments. Members asked to see the sample composition and the results before the numbers were fixed. Members of the General Board were asking for the figures. Nothing published suggests a firm in the pool will be told.

The published timetable runs calibration from June to October 2026, with the results benchmarked against national risk assessments. One step inside it deserves attention: the model output is compared with what supervisors already know.

Read that straight and it is sensible engineering. Read from inside a firm, it raises a question: could the weights end up tuned towards the view national supervisors already hold? If this page understands the method correctly, nothing published says how that is guarded against, and the supervisors’ own views are not published either.

What a firm can do

Do not build a board paper around a threshold you cannot cite. A paper going to your audit or risk committee may say that the firm is or is not likely to be selected. If so, it should say which numbers that conclusion rests on, and point to the instrument they come from. For eligibility, it largely can: the two figures above are published, though still in a draft. For scoring, it cannot, and the paper should say so rather than implying a precision that does not exist.

What can be done instead is to prepare for the data rather than the score. The selection exercise runs on information that firms supply through their national supervisors. Some firms can produce consistent customer counts by Member State, consistent transaction volumes, and a defensible account of how they count them. They are in a better position than a firm estimating its own score against weights nobody has seen.

And there is one question worth putting to your own supervisor now, in the ordinary course, while it is still a question about methodology rather than about your firm. Has it seen the indicator weights, and was it among the authorities whose knowledge the model has been calibrated against? The answer tells you how much the authority that supervises you today knows about the test that will be applied to you in 2028. That is a different and more useful thing than knowing the test yourself.

What it is, and where it stands

Central bank money in digital form, issued by the Eurosystem (the European Central Bank and the euro-area national central banks) and carrying legal tender status. It comes in two forms: online, used over a network, and offline, loaded onto a device and spent without a connection.

The Commission proposed it in June 2023. The Council agreed its position on 19 December 2025, Parliament’s economic affairs committee in June 2026, and plenary cleared the file into negotiations on 9 July 2026. Negotiations continued through the autumn. Parliament’s Legislative Train page for the file, as updated on 20 September 2026, described the holding limits as not yet discussed at political level.

Offline is treated like cash

Online payments are subject to the ordinary anti-money laundering rules. Offline is approached from the other end, on the footing that it resembles cash, and Article 37 sets out the consequence. Transaction data is not retained, by payment firms or by the central banks. What is retained is funding and defunding data: the amount loaded or unloaded, the device identifier, the date and hour, and the account numbers used. This data is to be made available to the financial intelligence unit and other competent authorities on request. The offline digital euro is therefore neither anonymous nor monitored. There is a record of money going on to a device and coming off it, and none of what happened in between.

The limit is the control

If nobody sees the transactions, the ceiling is what stops large sums moving unseen. Article 37 empowers the Commission to set offline transaction and holding limits, weighing the prevention of money laundering and terrorist financing against not unduly restricting use. The Council’s text states that specific limits for offline payments are essential to mitigate those risks. It is not a monetary figure with an incidental compliance effect but, offline, very nearly the only control there is.

Two limits sit in different hands. The overall holding ceiling is a Council decision on a European Central Bank recommendation. The offline limits, the ones doing the anti-money laundering work, stay with the Commission. Paragraph 8 lists four matters the Commission must weigh. The fourth puts a floor under the control: the limits must be high enough to allow purchases of necessities over at least 72 hours. A cap low enough to defeat crime may not be so low that somebody cut off from the network cannot buy food for three days. The 72-hour floor means the figure cannot be set very low.

Two paragraphs that create work

The Council added two provisions absent from the proposal. A Member State may ask the Commission to lower the offline transaction limit in its territory, on objective data showing that a lower limit is necessary. The Commission then adopts it for that country alone. And payment service providers must ensure their clients’ offline devices respect the limits applicable where the client resides.

Together they place national variation of an anti-money laundering control inside a Regulation. They also put the firm in charge of enforcing it, by knowing where each client lives and applying a different ceiling on the device. That is a build, and the numbers it would enforce do not exist yet. The same article obliges the Commission to request an opinion from the Anti-Money Laundering Authority (AMLA) and to consult the European Data Protection Board, both discretionary in the proposal.

Who sets the number is now the fight

Two press accounts, from the Italian newswire Public Policy on 28 October 2026 and from La Verità on 29 October 2026, report the negotiations stalled on the offline cap. The Council would have the figure fixed by implementing act through a committee of officials from the 27 Member States. Parliament wants it set in a form it can object to, which is what a delegated act gives, and, on the Public Policy account, based on a recommendation from the Authority. Both report that the Council’s route leaves out Parliament and the Authority, and that the two institutions agree the offline limit should sit below the online one.

The reports overstate one point, since paragraph 8 already obliges the Commission to request the Authority’s opinion. The real difference is whether the figure flows from a recommendation, or merely takes an opinion into account alongside three other factors. Neither account draws that line, and trilogues produce no documents to settle it. What the impasse does settle is which question matters. A delegated act can be blocked by Parliament or by the Council; an implementing act decided in committee cannot. The same number by those two routes is the same control with a different set of people able to refuse it.

What to do

Operationally, nothing. Three things are cheap. Know the shape, so that a standard landing later does not start from nothing. Watch both the limit and the instrument that sets it, since neither is settled. And make sure somebody in the business, not only in compliance, knows that an offline form exists and behaves like a loaded card rather than an account.

Outside the Union the direct reach is slight, this being euro-area money, though a European subsidiary providing payment services may end up distributing it. The wider point is that whatever the Union settles on will be quoted at firms elsewhere by their own regulators.

What Article 75 allows

Every bank sees its own customers and not what the bank down the road sees. The same customer can look entirely ordinary in each institution alone, and obvious the moment the two pictures are put side by side. Article 75 of the Anti-Money Laundering Regulation is the first serious European attempt at that. From the date the Regulation applies, members of a partnership for information sharing may share information among each other. They may do so only where strictly necessary to comply with their obligations under Chapter III, which is customer due diligence, and under Article 69, which is the reporting of suspicions. The sharing must respect fundamental rights and judicial procedural safeguards. Obliged entities are the core of such a partnership, and in defined cases financial intelligence units, supervisors and certain other public authorities may take part. It is permissive: members may share, and nobody is obliged to join.

Joining is a build, not a signature

The article sets most of it out already. Notify the supervisor. Before the partnership starts operating, the supervisor verifies that the compliance mechanisms are in place and that a data protection impact assessment has been carried out under Article 35 of the General Data Protection Regulation. That assessment is itself required before any personal data is processed. Record every instance of sharing. A partner’s information may not be relied on alone. It must be assessed before any decision affecting the business relationship or an occasional transaction is based on it. Where it feeds a decision to refuse or end a relationship, or to carry out an occasional transaction, the record has to show it came from a partnership. Sharing is limited to three groups of customers: those associated with a higher risk under the Union and national risk assessments, those falling under Articles 29, 30, 31 and 36 to 46, and those where more information is needed to work out the risk. Technical and organisational measures are expected, pseudonymisation among them, and information produced by artificial intelligence or an algorithm may be shared only where there was adequate human oversight. The internal policies have to be written before participation, not after.

The guidance arrives after the power does

On 1 July 2026 AMLA announced that it and the European Data Protection Board would jointly develop Guidelines on how these partnerships should be built, through a joint drafting team. It has announced the first of its roundtables to gather early views, and a public consultation on the draft, with a public hearing, is due in the first half of 2027. No adoption date has been given, and the reason for that is itself on the record.

At the General Board in joint composition on 1 July 2026, Members welcomed the plan and then raised concerns about the timing. The adoption date envisaged would fall after Article 75 had already started to apply. They encouraged AMLA to find a way to accelerate, and asked for consistency with related work at the EBA and the FATF. AMLA replied that effective guidance here requires close cooperation with the Data Protection Board, acknowledged the concern, and said it would explore expediting the work.

The conclusion records three undertakings, one of them unusually concrete. AMLA would remove the December 2027 deadline from the joint press release, so as to leave room for an accelerated process. No new adoption date has been given since.

Hold that against 10 July 2027, the day the power itself becomes available. On the announced timetable, the guidance on how to build one of these partnerships lawfully will in all likelihood arrive after the first day on which you are allowed to build one.

The first invitation shows what these calls look like. Published on 30 October, it gave applicants fourteen days, to 13 November, with the meeting on 30 November. It was addressed to technology providers rather than obliged entities, and capped at twelve, with one representative each. Participants were chosen against four stated criteria, with a scoring matrix behind them. Submitting an expression of interest, it says, does not guarantee participation. No equivalent call for obliged entities has been published, which makes it the thing to watch for.

Where strictly necessary is actually being decided

Article 75 permits sharing where it is strictly necessary. Everything a firm may lawfully do inside a partnership turns on that phrase, and it is not being settled in the consultation.

A working group was proposed to the Board in joint composition on 4 February 2026 by the Central Bank of Ireland and De Nederlandsche Bank. Their diagnosis is one of the clearest statements of the problem on the record. The article permits the sharing, but the drafting is high level, the terms are undefined, and the position on data protection oversight is uncertain. The group then split in two. A legal and regulatory subgroup took the scope and interpretation of the article, including what strictly necessary means. An operational subgroup took the verification process, timelines, notifications, amendments to partnerships and cooperation between authorities. An expert advisory group of financial intelligence units and non-financial sector supervisors worked alongside it.

Its final proposals were due to AMLA by the end of August 2026 and are to form a key basis for the Guidelines. So the meaning of the phrase that governs the entire regime is being worked out in a General Board working group, with its expert groups. The working group was proposed by two national competent authorities. All of this comes ahead of the consultation, on a draft for which the working group’s proposals are to form a key basis.

Why it is slow, and how fast it can land

These are Guidelines rather than technical standards, so AMLA adopts them itself under Article 54 of its own Regulation, with no Commission adoption and no step through the Official Journal. What follows instead is the two-month window in which every supervisor and financial intelligence unit has to say whether it complies. Once drafted, they can land quickly.

The delay is not indifference. Closing AMLA’s first conference in June, Vice-Chair Juan Manuel Vega Serrano said the private sector is eager to deploy its considerable resources more effectively. The hurdles, he said, are complex and not always fully harmonised data protection rules, which the Authority will have to walk with. Walk with, not remove. The Guidelines are being drafted jointly with the European Data Protection Board because the obstacle is not one AMLA can clear alone.

The benefit is bigger than this file

The new Payment Services Regulation was politically agreed in November 2025. The compromise text was confirmed by the Council’s permanent representatives on 22 April 2026 and by Parliament’s economic affairs committee on 5 May 2026. It decided against building a second sharing regime for payment fraud data, on the reasoning that partnerships of this kind already provide adequate frameworks. It goes further. Take a firm in a partnership established under Article 75 that lets it exchange the data needed to detect and prevent fraud. For that firm, the sharing done inside the partnership should be considered sufficient to ensure compliance with its obligation to share payment fraud data.

Two cautions. The instrument is not yet law: the Council adopted its first-reading position in late October 2026, Parliament's second reading is forecast for 14 December 2026, and publication follows that. The passages themselves are stable, though. They appear as recitals (147) and (148) of the Council's position, unchanged from the agreed text. The second caution is that they are recitals, which guide how the articles are read rather than standing on their own. The point stands all the same: the admission price of one partnership can answer two obligations, so a firm weighing that cost against a single benefit is weighing it against the wrong number.

The decision, and the lead time behind it

The decision has two parts, and firms tend to answer only the first. Do we want to share? And, if we did, could we? The second has a lead time. An impact assessment, a set of internal policies, a recording mechanism and a supervisory verification are not assembled in the month after a firm makes up its mind. For a group outside the Union this is a European instrument for European entities. But whether one joins is a group decision, and the data protection work will involve people who do not sit in the Union at all.

Nothing obliges a firm to join. The lead time is long enough that a firm which has not started asking the second question has effectively answered the first.

The test

Article 12(1) of Regulation (EU) 2024/1620 assesses credit institutions, financial institutions and groups of them operating in at least six Member States, including the home Member State. A Member State counts in one of two ways: through an establishment, or through services provided across the border above a materiality threshold. AMLA’s interpretative note of 12 May 2026, published with the reporting template, sets out both.

Establishments carry no threshold

Subsidiaries and branches count. So do sister undertakings that are credit or financial institutions and, less obviously, a network of agents where the network qualifies as an establishment. There is no volume condition on this route. A passported branch is an establishment rather than cross-border service provision, so six passported branches put a firm in the assessed population whatever its volumes.

Cross-border services count only above a line

Activity under the freedom to provide services counts where either of two limbs is met. One is more than 20 000 customers resident in that Member State as at 31 December of the previous year. The other is total annual incoming and outgoing transactions by those customers above 50 million euro. All of the activity counts, whether carried out directly or through branches, agents or distributors. On this page’s reading of AMLA’s materials, offering a product that a domestic customer can happen to use abroad, such as a credit card, is not operating in that country. The test appears to be active marketing to, and onboarding of, residents.

The group is added together

Where several credit or financial institutions in the same group provide services into the same Member State, their activity there is added together. Firms report one row for each entity and each country; AMLA does the aggregation and then tests the group. So an entity’s own figures can sit below both limbs in a country while the group crosses the line there, through a sister company it does not manage.

One point to check. AMLA’s frequently asked questions give the transaction threshold twice, in consecutive answers: twenty million euro in one, fifty in the other. Is the twenty an inconsistency? The standard and the interpretative note both give fifty, so fifty looks like the figure to work from.

What to do

The reference date is 31 December 2025 and the reference period the 2025 calendar year, so the first count is already fixed. It is something to find out. The work is to assemble customer residence and transaction volumes by country across every credit and financial institution in the group. Most groups hold that information. Few hold it in one place in that shape.

Two limits on it. The standard setting the thresholds is a final draft, not yet law. AMLA published its final report on 18 December 2025, and the Commission has neither adopted it nor published it in the Official Journal. The figures are firm enough to plan against, and not firm enough to rely on. And eligibility is not selection: the pool is assessed on residual risk. For the first selection, AMLA has said it will select up to forty, the cap that Article 106(2) sets for that round.

Three kinds of group should check. Those with offices in a few countries that sell remotely into others. Those with several small footprints in one country under different entities. And non-EU parents consolidating a European group, where the question lands on whoever consolidates.

The same control, nine months apart

Two banks read the Anti-Money Laundering Regulation and reach the same conclusion about a control they will need. The first builds it this autumn. The second waits for the final technical standards and builds it in May 2027. On 10 July 2027, when the Regulation starts to apply, both have the control written down and running, and on paper they are in the same place.

They are not. The first arrives with nine months behind it. The control has run, produced results, been found wanting in some respect and changed, and there is a record of all of that. The second arrives with a procedure and a go-live date.

The design can still be bought. A consultant writes the procedure, a vendor configures the system, and money compresses that work considerably. What money cannot compress is elapsed time.

Data is where it bites

A monitoring rule, a risk model, a threshold: each needs a history to be tuned against, and the history has to be captured before it can be used. Systems not recording it now will not have it later. That is the one item on the plan money cannot accelerate, which is a reason to start there rather than finish there.

Some clocks have already stopped. AMLA’s exercise identifying which institutions are eligible for direct supervision uses a reference date of 31 December 2025, covering the 2025 calendar year. Whatever the institution looked like last year decided the provisional pool, and the 2027 data collection confirms it. These processes look backwards, and the window that mattered is usually closed by the time a firm learns which one it was.

When anybody actually looks

The first selection of directly supervised institutions starts by 1 July 2027 and must conclude within six months. Direct supervision starts six months after the list is published, which puts the first supervisory cycle in 2028. By the end of 2027 AMLA expects to be 432 people, 200 of them recruited for direct supervision, working in joint teams alongside staff seconded from national authorities.

An examiner arriving in 2028 asks what the firm did and what happened when it did it. “We were waiting for the final text” is a true answer. It is not an answer to the question.

Build against the drafts

None of this needs a finished rulebook. Six of AMLA’s forty mandates had been submitted by July and three more went to the Commission on 1 October. Many of the resulting standards are likely to apply only after the Regulation does. Build against the drafts, record the assumption built on so that it is a decision on file rather than a drift, then let the control run and keep what it produces.

For a group outside the Union the same logic arrives through the European entity, which is the obliged entity. Much of what it will have to show in 2028 is produced by group systems and group functions sitting outside the Union entirely.

The design can still be bought. The track record cannot.

One framework means more reporting, not less

AMLA is building a central database. Under Article 11 of the Regulation that establishes it, the database is the single point at which supervisory authorities across the Union report and retrieve information. That covers information about obliged entities and about supervisors themselves. The standards specifying how and when supervisors report to it went out for consultation on 7 October, with a public hearing on 3 November.

A supervisor expected to supervise from data needs data. It is likely to ask for more than it asks today, more often, and in a form that matches what every other supervisor sends, because comparability is the point of building it. AMLA says it is drawing first on information supervisors already hold, automating the reporting where it can and taking care not to ask for the same data twice. That is plainly the intention. It does not change the direction of travel.

The cost is not really a compliance cost. It falls on systems, on the quality of the data in them, and on people who do not work in compliance at all.

The question to ask this quarter

Take any regulatory return the firm sends today. Who owns it? Not which department: which person, by name, can reconcile the numbers, explain them and sign for them.

That question did not matter much before, when a return of that kind was received and filed. It matters now because, for groups in the six-Member-State population, data the firm supplies feeds the assessment that decides whether it is supervised at home or from Frankfurt.

Volume is not the measure

There is a second half to this, about what is sent rather than how much. Rikke-Louise Ørum Petersen of AMLA’s Executive Board put it plainly in Copenhagen on 15 September: “More reports do not automatically mean better intelligence.” A report earns its place when it explains why the activity matters in the context of that customer. Recording that something looked odd is not the same thing.

Closing AMLA’s first conference in June, Vice-Chair Juan Manuel Vega Serrano pointed to a stark, collective challenge. Only a very limited number of suspicious activity reports, he said, actively feed into organised crime investigations. He added that this is a question of architecture rather than of will. Coming from the Vice-Chair of the new European authority, that is a notable statement.

It runs in both directions. In the same speech Ørum Petersen said institutions need to know what financial intelligence units consider useful. A firm being asked for better reports is entitled to ask which of its own were worth receiving, and that question now has support from the European side.

What to do

Budget for more reporting rather than less, and before anything else find out whose name is on it. For a group outside the Union, the European entity sits inside all of this. The data it reports, however, is produced by group systems that may sit entirely outside. The burden lands in one place and the work in another.

The rulebook is not weather

AMLA drafts the technical standards and guidelines in working groups and task forces with national authorities, which report to its General Board. Your supervisor is not waiting at the end of a conveyor belt for the output; it is in the room where the output is made, and it takes positions there. So when a draft creates a real operational problem there is somewhere for it to go. The instinct is to absorb it and plan around it. The alternative is to write it up and take it to the supervisor, with the detail of what the requirement does to systems, customers or timetable. Then ask for it to be carried into the working group. That costs nothing beyond the writing.

Where the doors are

AMLA opened a consultation on the standards for its central database on 7 October, with a public hearing on 3 November and written comments closing the same day at four in the afternoon. It is currently the only one open. Roundtables on simplified customer due diligence run in Frankfurt from 9 November to 2 December, with non-financial firms specifically encouraged.

Respond directly where the firm has a distinct position, and through a trade association where the industry has a common one. A trade body response can show that a concern is shared across an industry, which separate letters may not. Membership of a European body, not only a national one, matters for a practical reason. Before consulting on the non-financial risk methodology, AMLA held a stakeholder roundtable in May with limited places. The invitations went to EU-level associations representing non-financial obliged entities, not to national bodies or to individual firms. Two obstacles are also smaller than they may look. Submissions may be made in any official language of the Union. And on several consultations this year, AMLA has said in terms that it particularly encourages feedback from the non-financial sector.

Attending is not responding

Attending is not responding. The first public hearing, on the customer due diligence standards in March, drew more than 1 600 participants, and the larger hearings since have drawn a thousand or more. Written responses arrive in dozens or low hundreds. AMLA said after that first hearing that what was said would supplement the written contributions rather than stand in for them. The written submission is the part that enters the file.

AMLA has said what a useful response looks like. Speaking in Frankfurt in May, Executive Board member Rikke-Louise Ørum Petersen asked for detailed, experience-based and data-driven input. She asked for specifics on where definitions lack clarity, where implementation creates friction, and where requirements could be made more operational, or simpler, without lowering standards. That is a brief, not an invitation to call a requirement burdensome. It means naming the words in the paragraph, what the systems do when they meet them, what it costs, and what wording would work instead.

One case where it worked

The standards on enforcement grade a breach of the anti-money laundering rules and decide what follows from the grade. The European Banking Authority drafted them and consulted in spring 2025. AMLA consulted again to ensure that the non-financial sector’s views were fully captured. It put the same draft out from 9 February to 9 March 2026, for the stated purpose of hearing from that sector. Ninety-one responses came in, eighty-eight of them valid, from more than twenty Member States, and seventy per cent were non-financial.

The text changed. Articles 1, 2 and 4 were amended so that the standards work for the non-financial sector as well as the financial one. A criterion on financial strength was also widened. A supervisor can now use any reliable information it holds, rather than the regulatory capital figures only a bank would have. The standards were finalised on 8 July 2026.

Who does turn up

The group-wide consultation drew 123 responses, every one of them published. Fifty-two came from the non-financial sector, among them notaries, bar councils, accountants, art dealers, watchmakers, casinos, estate agents and two football bodies. This is not a conversation between AMLA and the banks.

One sector is largely absent. Asked in April whether football had engaged, the Chair answered that things were “perhaps a little quiet on that front”. Football representatives at her roadshow roundtables, she said, were “more the exception than the rule”. That matters because a methodology can only reflect the evidence put to it. A market that does not put its case may have its settings chosen by the markets that did.

The second channel

In October 2025 the Regulation establishing AMLA was amended. It now gives the Authority the task of reviewing how adopted technical standards work in practice, and of proposing changes. Those expressly include changes to remove redundant or obsolete reporting requirements and to make reporting proportionate and consistent. When it gives an opinion on legislation already in force, it may say whether such a requirement ought to go, and consult all relevant stakeholders on that question. The Commission may then put a proposal to Parliament and the Council. No consultation has been opened under it yet.

The first channel is for a rule still in draft, the second for a rule already law that is not earning its cost. Closing AMLA’s first conference in June, its Vice-Chair asked the room to challenge the Authority wherever necessary so that it can refine its processes. The invitation is not the difficult part.

A group outside the Union has no supervisor in those rooms, but its European entity has one. The problem that entity raises with its own supervisor is the group’s only route into the drafting.

The four

On 1 October AMLA finished four sets of regulatory technical standards. Three went to the European Commission the same day; the fourth, on how supervisors work together across borders, was published as a final draft. If the term is doing a lot of work there, the short explainer below sets out what these are and why they bind rather than advise.

The first, under Article 19(9) of the Anti-Money Laundering Regulation, answers when a customer is actually a customer. It sets the criteria separating a business relationship from an occasional transaction, explains how to identify linked transactions, and covers lower thresholds. It is the least discussed of the four and probably the one most firms will feel first, because it decides when full customer due diligence is triggered at all.

The second is the customer due diligence standard under Article 28(1). It covers what must be collected and what verified, proportionate measures for lower risk, remote verification and electronic identification, and screening of politically exposed persons. The third sets minimum requirements for group-wide arrangements under Articles 16(4) and 17(3), including what happens when third-country law stops a subsidiary doing what group policy requires. The fourth, under Article 46(4) of the sixth Anti-Money Laundering Directive, is about supervisors rather than firms. It covers home and host roles, joint cross-border inquiries, and one EU supervisor passing information to another without obtaining consent first.

The group-wide standard is probably yours

One point of scope is easy to get wrong in the direction that lets a firm off. The group-wide standard is tempting to read as reaching only groups with operations outside the Union. It does not. Article 16(1) and (2) require a parent undertaking to make its internal procedures, risk assessment and staffing requirements apply in all branches and subsidiaries in the Member States. It must also run a group-wide risk assessment and group-wide policies, and establish compliance functions at group level. A wholly European group is squarely inside that. What does depend on having something outside the Union is the rest. The third-country machinery in Article 17 is triggered by a branch or subsidiary in a third country. The rules on which European entity becomes the parent apply where the head office sits outside.

Three situations outside the Union

A bank inside a European group is reached by the group-wide standard through its parent, which must apply group policy to it. Where local law prevents that, AMLA sets out a sequence. Identify the impediment, tell the supervisor what the local law stops you doing, seek the customer’s consent where legally possible, then apply additional measures. Those are the familiar ones, from restricting the business to lower-risk products through to closer monitoring. What is new is the end of the sequence. If the risk still cannot be managed, the supervisor can require the group to stop taking on new business, end existing relationships, or close some or all of its operations in that country. That power sits in Article 17(2) of the Regulation itself, not only in the standard. Where the home supervisor considers the additional measures insufficient, Article 17(2) says it “shall exercise additional supervisory actions”, including requiring the group not to enter into business relationships, to terminate existing ones, not to undertake transactions, or to close down its operations in the third country. AMLA’s factsheet on legal impediments also covers this.

A bank with its own establishments in the Union holds obliged entities, and the rules apply to them directly. If it has more than one, what it tells a supervisor in one Member State can be passed to the others without anyone asking first. Where the only link is correspondent banking, nothing here binds the bank directly. But the due diligence standard sets what the European counterparty must collect and verify about it, which is why the questionnaires change.

The gap nobody is planning for

Inside the Union, the first two are the ones to read first, unless the group has operations outside it. The classification standard decides how systems label a relationship, and the due diligence standard decides what onboarding has to capture.

Then the arithmetic. Under Article 49 of the AMLA Regulation the Commission has three months to decide on a draft standard. Under Article 51 Parliament and the Council then have three months to object, extendable by a further three. Publication in the Official Journal and entry into force follow, and AMLA proposes that the three standards for firms apply six months after that. The supervisory-cooperation standard is drafted to apply from 10 July 2027. Run it from 1 October 2026 without a single delay and the second half of 2027 is the earliest, with early 2028 just as likely. The Regulation itself applies on 10 July 2027.

So on the morning of 10 July 2027 firms will be applying Articles 19, 28, 16 and 17, and the standards explaining how will most likely not be in force. That could change if Parliament and the Council say early that they will not object. It is not something to build a programme on.

The timing reflects the scale of the task. AMLA is building an institution and writing a rulebook at once, with a few hundred people, in public. On 2 December 2025 its Chair told the European Parliament: “We are, as it were, flying a plane while building it.” The plane is still being built and the date has not moved.

What to do

The drafts are public, and they are the best statement available of what the Authority expects. Read them as drafts and build to them. Treat the gap between the two dates as a problem to manage rather than a grace period. The obligations in the Regulation apply from 10 July 2027 whether or not the standards are in force by then. For football agents and professional football clubs the standards apply only from 10 July 2029. These four are among the first of around forty mandates, and what is still owed and when it falls due is on the mandate page.

The list decides who holds the file

In July 2026 the Chair of the Anti-Money Laundering Authority (AMLA), Bruna Szego, appeared before two committees of the European Parliament. Most of the attention went to the forty or so firms AMLA will supervise directly from 2028. The point that mattered more was that AMLA is building a common supervisory methodology and handbook together with national supervisors. The aim is that supervision works the same way across the Union, whoever is carrying it out.

At the hearing on 15 July 2026, Ms Szego explained that the model combines inherent risk and the quality of controls to produce a residual risk score.

That is the structure of a business-wide risk assessment. What changes is who does the scoring, and what follows from it.

What changed since the summer

Nothing dramatic, which is the point: no new directive and no emergency measure. AMLA simply began behaving like a supervisor. It had 157 staff on the day of the hearing and expects 233 by the end of this year. On 1 October it finalised four technical standards, including the central one on customer due diligence. The Regulation and the Directive set 10 July 2026 as the date for submitting all four. Three went to the Commission the same day. The fourth was published as a final draft. The difference matters, because under Article 49(1) of the AMLA Regulation the Commission’s three months run from receipt of a draft, not from its publication. It consulted on several more, and asked the Commission to look again at crypto-assets, including staking, lending and decentralised finance.

Two developments from outside the headlines

The first is a change of method rather than of degree. In January 2026 the Union added the Russian Federation to its list of high-risk third countries on its own assessment, rather than following a decision of the Financial Action Task Force (FATF). This created a new category in the annex for countries whose FATF membership is suspended. Russia is under no call for action and no increased monitoring, so this is the Commission listing a country on its own reading of the evidence. It is the first listing of that kind to take effect, and it is a signal about how listings may work in future.

The second is the Council of Europe protocol on asset recovery, signed by the Union on 14 October. It gives asset recovery offices fourteen days to answer a reasoned request about accounts, or three days where the request is urgent. The qualifier has travelled less widely than the three-day figure. Those limits apply only to databases the office can reach directly. They stretch to twenty-one days and seven where the information is available only indirectly or answering would impose a disproportionate burden.

Three dates follow. AMLA scheduled a public hearing on the standards for its central database for 3 November. Member States were required to transpose the asset recovery and confiscation Directive by 23 November 2026. It obliges each to run an asset recovery office with access to beneficial ownership registers. The first selection for direct supervision must begin by 1 July 2027.

Why waiting stopped working

The instinct, faced with a rulebook still being written, is to wait for the final text so as to implement something rather than interpret a draft. It is a good instinct, and the one every well-run compliance function has been rewarded for. Under the old regime waiting still left time to build.

It does not any more, because the two clocks have come apart. The application date is fixed. The standards meant to explain it are not. Waiting no longer buys certainty; it buys delay and calls it prudence, and a first real attempt at implementation then lands on the wrong side of the deadline.

So the framework reaches a group outside the Union through its European branch, on a date fixed in the Regulation, by a method still being written. The text is not final. The deadline is.

There is no big bang

Article 26 of the Anti-Money Laundering Regulation sets no date on which the existing book must be compliant in full. It sets a cadence. The period between updates depends on the risk posed by the business relationship. But it shall not in any case exceed one year for higher-risk customers, meaning those to whom enhanced due diligence applies, and five years for everyone else.

The frequency is the firm’s to set, on risk. The outer limit is not. The obligation is rolling, with a backstop, and the work spreads across years rather than piling into one.

Recital 76 is frank about why the caps exist. The existing approach is already risk-based, but given the higher risk associated with certain intermediary structures it might not allow for timely detection. It is therefore important, the recital says, to ensure that clearly specified categories of existing customers are also monitored on a regular basis.

What brings a review forward

Article 26(3) requires a review, and an update where relevant, in three cases. The first is a change in the relevant circumstances of a customer. The second is a legal obligation during the calendar year to contact the customer, in order to review beneficial ownership information or to comply with the Union’s tax cooperation rules under Council Directive 2011/16/EU. The third is becoming aware of a relevant fact about the customer. The middle one ties the review to contact the firm already has to make. A firm that runs its tax reporting and its anti-money laundering refresh as unconnected exercises is paying twice for the same letter.

Sanctions run on a separate clock. Under Article 26(4) a firm must regularly verify whether the customer or the beneficial owners are subject to targeted financial sanctions. The frequency must be commensurate with its exposure to the risk of non-implementation and evasion. Credit and financial institutions must also do it on every new designation. That duty neither waits for the five-year cycle nor is satisfied by it. The same paragraph says it does not replace the obligation to apply sanctions, or any stricter screening requirement under other Union or national law.

The draft standard adds a trigger the firm does not control. The firm must determine whether existing customers and their beneficial owners are or have become politically exposed persons, family members or close associates. It must do so at a risk-based frequency, and without delay when the list of prominent public functions changes. That is smaller than it sounds, because under Article 43(5) the Commission assembles the national lists into a single consolidated one, published in the Official Journal and on AMLA’s website. The standard (Article 17(3)) allows automated screening tools, manual checks or a combination, provided they are effective and proportionate to the firm’s business, risks, complexity and size.

When the clocks start

The technical standard on customer due diligence ends with a transitional provision. Some business relationships will have existed before the standard enters into force. For those, the documents, data and information are brought in line with the standard and the Regulation on a risk-sensitive basis, within the periods set out in Article 26(2). The existing book comes up in risk order, with one year and five years as the backstop, and there is no remediation deadline.

A recital most readers will never reach is worth more than the article. For relationships existing before that date, the maximum periods of one and five years “should only start with the date of entry into force of this Commission Delegated Regulation”. For a legacy customer the clock runs neither from the last time anyone looked at the file nor from 10 July 2027.

Then the two provisions read together. Entry into force is the twentieth day after publication in the Official Journal, and application is six months after that. So the legacy clocks start about six months before the standard itself begins to apply. A low-risk legacy relationship has five years from entry into force, which is roughly four and a half years from the day the obligation actually bites. AMLA refused to extend the one-year and five-year periods, which it could not have done in any event, since they belong to the Regulation rather than to the standard.

The duty starts before the detail arrives

Article 26 applies on 10 July 2027 because the Regulation does. The standard that fills it in applies six months after entering into force, so it most likely starts some months after the Regulation it supplements. Its sister standard on business relationships and linked transactions was drafted to apply from 10 July 2027 flat, and that did not survive either. The guidelines on ongoing monitoring are in a worse position again: Article 26(5) required AMLA to issue them by 10 July 2026, and they have not yet been issued in final form.

So for a period firms will be operating a cadence obligation under the Regulation alone, without the standard explaining what a review has to contain. That is a reason to be able to show the reasoning, because for a while the reasoning is all there will be.

The work worth doing now

The question is not whether the files are good. It is whether the firm can produce, for every customer, the date of the last review and the current risk classification. A low-risk file cannot be shown to be inside five years if nobody recorded when it was last touched. A higher-risk file cannot be shown to be inside one year if the classification lives in somebody’s judgement rather than in the system. Take a book of 40 000 relationships. The review date sits in a spreadsheet for retail customers and in the core system for corporates, and the two disagree on about a tenth of the file. Nothing in that firm’s controls is wrong, and it still cannot answer the only question the cadence rule asks. That is a data exercise rather than a compliance one, and no version of the final standard can make it unnecessary.

One group reads all of this differently. For a firm newly in scope, every customer is an existing customer with no file at all, so the risk-based transition is not a concession but the whole shape of its 2027. Work out which relationships would attract enhanced due diligence, start there, and record why.

What the standard decides

A suspicious transaction report goes to the financial intelligence unit in the filer’s own country, and sometimes the matter belongs elsewhere. Article 31(1) of the sixth Anti-Money Laundering Directive says that where a report concerns another Member State the unit must forward the report, or all the relevant information obtained from it, promptly. Promptly is the only time standard in the text. That leaves the harder question of when a report concerns another Member State. Each country has answered it in its own way. On AMLA’s view, the result has been delay, legal uncertainty, and both over-sharing and under-sharing. Article 31(3) tells it to write the answer down, and the draft consulted on from 6 July to 6 October 2026 is that answer.

It builds two routes. In the first, a cross-border report, the whole report is sent on to the other country’s unit. This is for the clear case: a firm serving a customer in another Member State under the freedom to provide services, without being established there. In the second, a cross-border dissemination, only the relevant information is sent. It catches less obvious links, such as a subject of the report, or assets connected to it, being in another Member State. Both can happen at once.

The choice at the heart of the draft is who decides relevance. The sending unit applies only the selection criteria and does not judge whether the material matters to the other country. The receiving unit does that, against its own criteria and national risk assessment. If it wants the rest, it asks for it through FIU.net. The unit that understands its own country decides what is useful to it.

The part that concerns a firm

The criteria had to be mechanical, because AMLA wants them applied automatically inside FIU.net. Mechanical means they read fields. For a natural person the first test is usual residence, then nationality, then country of birth. For a legal person it is where the business takes place, then the registered address or official office, then the country of creation. Before anything can be built at all, the draft asks for minimum identifying data. For a natural person, that is at least one of an account number, a date of birth, an identity document number or a residence. For a legal person, it is at least one of an account number, a registration number, a VAT or tax number or a registered address.

So the plain question is how often the residence field is actually filled in, or the date of birth, or the registration number of a corporate customer. If those fields are empty or wrong the test may find nothing, and the report may not reach the country it concerns. The press release is right that the standard imposes no obligation on firms. What it does is give an obligation they already have a consequence it did not have before.

There is a second edge to the same point, and AMLA put it on the record rather than in the consultation paper. AMLA was asked at the Board in financial intelligence composition about the practical use of matching tools in cross-border analysis. It said the draft requires a full match for certain selection criteria. It also acknowledged that the point needs further consideration within the mandate. A full match is a demanding test for a firm’s data. A name or an identifier recorded loosely does not match, and what does not match does not move.

The same meeting recorded that disseminations are tiered. At the first stage only a limited set of key information goes across, and the receiving unit decides from that fragment whether to ask for the rest. So whether your report travels in full can turn on how much of it survived the first test.

The companion standard decides the fields

A separate standard, out for consultation in the same summer, sets one set of data points for the whole Union, completed according to activity and the nature of the suspicion. They come in five grades rather than two. Some are simply mandatory, and within those a technically required subset will stop a report being submitted at all if missing. Others are mandatory if you have them. For those, the state of a firm’s own records decides the quality of the report. The rest are optional, or required only where something else applies, or where a national authority asks.

The accompanying technical note runs to 167 numbered data points and 42 kinds of link between them, so the form is larger than the word template suggests.

A number and a date

The number is three. A dissemination is triggered by a connection to a primary subject, and one way of being one is a financial ranking of three or higher. The ranking runs on a scale of one to five, where three means a share between 25 and 50 per cent. It takes the higher of two shares: the subject’s share of the transactions, or its share of the total transaction value. A subject accounting for a quarter of the money in a report is, on this test, a reason to tell another country.

The date is 10 July 2028, when financial intelligence units are to have implemented the hit and no-hit matching process inside FIU.net, a tool whose name is written Ma³tch. Under this draft a full positive match is itself a reason to send information on, so that is when one of the routes starts working properly. It comes from the consultation paper rather than an adopted text, so treat it as a plan rather than a promise.

What the subject can see

One question hangs over all of this: what the person in the report may later find out. It is before the Court of Justice as Case C-222/25, a reference from the Estonian Supreme Court on which data protection regime governs financial intelligence units.

The Advocate General opined on 8 October 2026 that the Law Enforcement Directive applies. On that view, a unit may restrict access, and even refuse it without giving reasons, where safeguards exist and there is an indirect route through the supervisory authority. Judgment is to follow.

Two cautions, and they matter. The Court’s press release on the Opinion is an unofficial document for media use. And an Opinion does not bind the Court. Read it as the direction the case is travelling in rather than as the answer.

What to do about it

Ask for a count. Take last year’s suspicious transaction reports and find out how many went out with residence blank, with no date of birth, with no registration number for a corporate subject. If the answer is reassuring, nothing follows. If it is not, there is now a number to take to whoever owns the quality of the reporting data, and a reason that is easier to explain than most.

This is still a draft and the consultation results are yet to come. If adopted as drafted it would apply from 10 July 2027, the day the Regulation itself applies. Cross-border reports would use the template in a separate implementing standard. AMLA finalised it and sent it to the Commission in July 2026, so the form exists as a final draft, though the Commission has not yet adopted it. What is still being settled is the rule about when it has to be filled in.

For a group outside the Union, nothing here is filed and no European unit is reading its reports. But a European entity serving customers in other Member States under the freedom to provide services is exactly what the first route was built for. Its reports are designed to travel.

The problem it solves

European law wants group-wide policies, procedures and controls, and a group-level view of risk. That assumes somebody sits at the top of the group inside the Union, and often nobody does. A group run from New York, Tokyo or Zurich may have two, five or fifteen obliged entities in the Union. None of them is the parent of the others, and each answers to a head office outside it.

The standard on group-wide requirements answers the question directly: in that situation, which European entity is the parent undertaking for these purposes? The answer is not left to be minuted internally.

The three tests

The first is sufficient prominence in the Union. Any one of four factors may determine it, rather than all of them together. Whether the entity is a holding company of the listed kinds at the highest level of consolidation, provided the third-country head office established it in the Union for most of the group’s obliged entities there. The number of cross-border establishments, counting agents, distributors and arrangements under the freedom to provide services. The average number of customers over the two preceding calendar years, and in particular those classified as high risk. The average of incoming and outgoing transactions over the same two years. Where those do not settle it, three further factors do: total annual turnover, balance sheet size, or the average number of employees during the financial year.

The second is sufficient understanding of operations in the Union. It is shown first by qualitative criteria on the powers and influence an entity actually has over the others. If those do not settle it, the number of full-time equivalent staff in the compliance function decides. The third is whether the entity actually has responsibility for implementing the group-wide requirements.

Those three come from Article 2(1)(42)(b) of the Regulation itself, which applies where at least two subsidiary undertakings in the group are obliged entities established in the Union. The same provision requires the entity chosen to be an obliged entity and not the subsidiary of another obliged entity established in the Union. What the standard adds is the content of each test.

The drafting history matters here. The consulted draft ran prominence as a strict waterfall over three years, and provided that where the two tests pointed at different entities prominence prevailed over understanding. AMLA says it revised the criteria in light of consultation feedback. Three things changed: the waterfall became a set of factors, the look-back fell to two years, and the rule that prominence beat understanding was deleted. The choice now rests more on operational control, decision-making authority and capacity to ensure compliance, which is judgement to be exercised and recorded rather than a formula to apply.

Note what the second test counts. A group that has spent five years consolidating compliance into one European hub, for sensible reasons of cost and consistency, may already have chosen its parent undertaking without anybody deciding to. That is not a flaw in the drafting, since counting compliance staff is a reasonable proxy for which entity understands the group’s European operations. It is worth knowing that an organisational decision taken for other reasons has this consequence.

It is an approval, not a filing

The entity identified as parent must notify its supervisor, and the notification has to carry six things. An overview of the organisation, roles and responsibilities of the identified entity and of every other obliged entity of the group in the Union. A detailed chart of the group, with the shareholdings in capital or voting rights, and an overview of what each entity does and where. Detailed reasons for the determination of sufficient prominence. Detailed reasons for the determination of sufficient understanding of operations. Detailed reasons that the responsibility for implementing group-wide requirements is met. And anything else needed to evaluate the identification.

The supervisor confirms receipt, tells the supervisors of every other obliged entity of the group in the Union, and tells AMLA. Where the supervisor is a self-regulatory body it also tells the public authority overseeing it. Then the operative words: the supervisor shall approve the identification within 60 calendar days of receiving the complete notification. This is a submission on which a decision is taken, circulated to every supervisor with an interest in the group and to the European authority, with a clock attached.

The day-one problem

A group already in this situation on the day the standard applies must notify on that date, not within a grace period afterwards. A group arriving in the position later has twenty-eight calendar days. The clock starts on the day two or more obliged entities of the same non-European head office end up established in the Union without being in a parent and subsidiary relationship. A group already there on the start date gets none of those days.

So the analysis, the group chart, the three sets of detailed reasons and the internal agreement about which entity takes this on all have to be finished before the start date. That date is not yet fixed. AMLA dated the final report 30 September 2026 and published it on 1 October 2026. The standard now sits with the European Commission. It applies six months after it enters into force, which is itself twenty days after publication in the Official Journal. On a reasonable reading that puts application in late 2027 at the earliest, and quite possibly in 2028.

What the parent then owes

The parent maintains an organisation and coordination structure at group level with real decision-making powers. It carries out the business-wide risk assessment at group level and keeps it updated. It makes sure information moves between functions, and between group and local level. It identifies and mitigates conflicts of interest, and it makes sure the management body and control functions have what they need.

It also carries the third-country regime. Where the law of a country outside the Union blocks the European rules, the parent has 28 calendar days to tell the home supervisor which country and exactly how. It must make its branches and subsidiaries work out whether customer consent can lawfully overcome the obstacle, and require that consent so far as local law allows. If that fails, it chooses from twelve additional measures. The last two are terminating the relationship, and closing down or divesting some or all of the operations concerned. These responsibilities fall on the entity identified under the three tests, and compliance staff numbers are one factor in that identification.

What to do now

Four things, two of which cost nothing. List the group’s obliged entities in the Union, which is a different and usually shorter list than its legal entities. Run the three tests on them and write down the answer with reasons, which will be needed in a notification eventually and tends to surface disagreements better had early. Count the compliance staff in each, because if the tests point somewhere the group would not have chosen, that is a decision to take deliberately rather than discover. And start the group chart, which is not a document most groups have lying about in the form a supervisor will want.

One footnote for anyone not in a group at all. Networks, partnerships and franchises are treated as structures rather than groups. The final text also added a route the consulted draft did not have. Where no entity holds sufficient responsibility and powers, a coordinating obliged entity may ask permission not to identify a head of the structure at all. For a structure, somebody does not necessarily have to be the parent. For a group, somebody does.

Two provisions to check first

The first decides who the customer is. For most obliged entities that question answers itself. Article 19(6), point (c), of the Regulation answers it for estate agents, and the answer is both parties to the transaction, not the side that instructed. The same rule applies to a notary, a lawyer or another independent legal professional intermediating a transaction, where they are the only one doing so. An onboarding process built on the vendor, or on the buyer who walked in, is missing half the file.

The second decides when letting is caught. Article 3(3)(d) brings in estate agents and other real estate professionals to the extent they act as intermediaries in real estate transactions. It says expressly that this includes letting where the monthly rent is at least ten thousand euro, irrespective of the means of payment. That last phrase is deliberate. A lettings business that placed itself outside the framework because nothing was paid in cash was reasoning from the old habits of the subject.

What AMLA published, and what it is not

In a thematic note dated 9 June 2026 and published in July, AMLA set out its view of money laundering and the real estate sector. Its first page describes it as a technical overview of academic literature and policy reports, intended to inform discussion. It states that it does not constitute a policy position of the Authority or its governing bodies. It binds nobody and commits AMLA to nothing.

It is still the clearest account available of how this sector’s future supervisors understand it, and it carries figures with sources attached.

Its first finding lists what makes the sector attractive: size, stability, economic returns, frequent opportunities to conceal beneficial ownership, lack of transparency in asset valuation, and decentralised transaction processing through professional intermediaries. In this first finding, the agents and the notaries appear among the features that make the sector attractive, not as the solution to it.

The argument that turns this political

Economic models predominantly predict two effects. Laundering raises house prices, because it increases demand and reduces supply. And it misallocates capital, because assets get chosen for their capacity to obscure rather than for their return. The empirical evidence is thinner, and the note says so: measuring laundering is extremely hard, and proving it caused a price movement is harder still.

The figures it cites are these. Money laundering is estimated at between one and six per cent of gross domestic product. In the United Kingdom, offshore ownership accounts for fifteen per cent of top-end properties and has been estimated to raise local prices by up to eight per cent. A Canadian policy report puts five per cent of transactions there as laundering-motivated, with a price effect between 3.7 and 7.5 per cent. AMLA cautions that the study rests on numerous assumptions which do not extrapolate easily. A small number of studies find a positive correlation between proxies for laundering, suspicious transaction report volumes among them, and local house price indices.

No single figure there will bear much weight. The link itself is what matters. Once a regulator connects laundering to housing affordability, the subject becomes a political question as well as a technical one. Will that affect how closely the sector is supervised? The note makes the link explicitly, observing that housing is a necessity good and that distortions limit economic mobility and the movement of workers between regions.

Where the suspicion sits

For real estate, AMLA says, the framework means “looking beyond onboarding checks and assessing how a transaction is structured, who is involved, who ultimately benefits, where the funds come from, and whether the commercial rationale remains credible as the deal progresses”. It adds that suspicion in this sector may appear in payment flows, but may also emerge through changes in ownership, funding arrangements, intermediaries, timing, urgency, pricing, or the wider structure of the deal.

That describes what an agent observes rather than what an agent handles. The buyer who appears two weeks before completion in place of the one who made the offer. The price agreed well above or well below anything the street supports. The sudden urgency, the funding that changes character, the intermediary introduced late with no explanation of what they do. The person who never touches the funds is often the only one in the chain who watches the deal take shape from beginning to end.

Three words in that quotation do most of the work: as the deal progresses. A commercial rationale that was credible at instruction and is no longer credible at completion is a change the agent is expected to notice. The obligation runs through the life of a transaction rather than attaching to its start, and a file of onboarding documents does not discharge it.

Two things aimed at somebody else

The note names six promising policy approaches, and one of them points inward: strengthening the accountability and public oversight of self-regulatory bodies. In the legal professions alongside this sector, a self-regulatory body is often the supervisor. AMLA’s own research lists strengthening the accountability and public oversight of such bodies among its promising approaches, which suggests the direction supervision may travel.

The other is structural. A firm operating under a franchise, or belonging to a network, should look at the group-wide requirements. The final standard treats a structure as two or more obliged entities sharing common ownership, common management or common compliance control. Any one of the three is enough, and compliance control is not limited to anti-money laundering compliance. AMLA’s own recitals name real estate franchises as an example. Separately, once a network has establishments in at least two Member States other than its home one, a supervisory college can be convened over it. A franchised brand that thinks of itself as a group of independent businesses with a shared sign should take advice on that before 2027 rather than after.

What to do

Check that the customer file covers both parties rather than the side that instructed. Check whether any lettings cross the ten thousand euro monthly threshold, cash or no cash. And record, for each live transaction of any size, what the commercial rationale is, so that when it changes somebody can see that it changed.

The deadlines and the data behind all of this sit in the implementation handbook. The thematic note itself is short, free and worth reading in full.

Nothing passing through the account was never the test. What the agent could see was always the test. AMLA’s thematic note, which states that it is not a policy position of the Authority, now describes the signs an agent is placed to see.

What changed

AMLA consulted on the standards on group-wide requirements between 16 April and 15 June 2026, and published a final text that is shorter, softer and far less prescriptive than the draft. Twenty-seven articles became twenty. AMLA lists 123 respondents.

Four things changed. The first is which entity carries the group obligation. In the draft this was a waterfall. A holding company came first. Failing that, the test was the higher of a customer count over the previous three years or the average of incoming and outgoing transactions over the same period. Failing that, it was the highest total turnover. The final text replaces it with a set of factors, any one of which may settle prominence. The look-back falls from three calendar years to two. The tie-breakers widen from turnover alone to turnover, balance sheet size or average employee numbers. And the rule that prominence automatically beat understanding of operations has been deleted.

The second is the group-wide risk assessment. Respondents read the draft as demanding that every entity’s individual risk profile be assessed and consolidated, however irrelevant. The final text asks instead for a holistic group-wide view, focused on the risks with a significant impact on overall exposure.

The third is third countries. The draft named five legal impediment scenarios, each with its own prescribed measures. The scenarios are gone, replaced by one common set of minimum actions plus tailored additional measures where needed.

The fourth is proportionality. A broad range of respondents asked for an overarching clause. AMLA declined, holding that proportionality works better written into each requirement where it is relevant.

Why three of those cost more

Under the draft’s waterfall nobody had to justify the outcome, because the test produced it. The steps were applied and the answer fell out. Now the factors sit at the same level and have to be weighed, across three tests: sufficient prominence, sufficient understanding of operations, and responsibility for implementing the group-wide requirements. The notification to the supervisor has to carry detailed reasons on each of the three. It also needs a detailed group chart with shareholdings in capital or voting rights, and an overview of what every European entity does and where. Anything else the supervisor needs to evaluate the identification goes in too. Where a sum would once have been produced, a reasoned case now is. The supervisor then has sixty calendar days to approve it, and that clock starts only when the notification is complete.

Proportionality runs the same way in reverse. Had the overarching clause been granted, a firm could have pointed at one provision and argued proportionate treatment generally. Because it was refused, the argument has to be made provision by provision, and recorded every time. More defensible, and more work.

Then the legal impediments. Matching facts to one of five named scenarios and taking the listed measures is an exercise a competent junior can run. Choosing from a set of tailored measures, and being able to say why those and not others, is not.

The one real saving

The group-wide risk assessment is a genuine win. An assessment that scaled with the number of entities would have been very large. One that scales with the number of material risks is smaller and more focused. For a group with forty European entities, most of them small, the final text asks for a much narrower assessment.

The shape is the same even here, though. Materiality is a judgement. A firm that leaves twenty-eight entities out of its group-wide assessment has to be able to explain why those twenty-eight have no significant impact on overall exposure. The work that went was mechanical. The work that replaced it is smaller and harder.

Where nothing softened

The list of additional measures available where a third country blocks group-wide policies grew from nine in the draft to twelve. The last two are termination of the relationship, and closing down or ensuring divestment of the operation. Respondents asked for thresholds or small-scale exemptions for smaller entities inside networks and franchises. AMLA did not accept this, on the view that exemptions would defeat the point of extending group-wide requirements to those structures at all. The twenty-eight day deadline for notifying a third-country legal impediment did not change either. Some respondents proposed a two-step notification modelled on the operational resilience rules.

AMLA’s reason is that twenty-eight days is not new. The same period has applied since 2019 under the existing delegated regulation on third-country impediments, so a firm with a working notification process under that instrument already owns the machinery. What the industry got instead of time was consolidation. Several impediments can now go into one notification. In a group the parent files, with the option of letting individual entities notify their own supervisors where a single notification would not work.

Who flexibility suits

Flexibility is not neutral between firms. A well-resourced compliance function prefers judgement to a formula, because a formula fits nobody exactly and a judgement can be tailored to the business. A thinly staffed one is better off with the formula, because one person with a spreadsheet can operate it. A documented judgement needs someone who can write the argument and someone else with the standing to sign it.

Does a framework built on documented judgement suit firms with the capacity to write that judgement down better than it suits smaller ones? If it does, that is not necessarily the wrong result. But a smaller firm should plan for it.

What to do

Build the file, not the spreadsheet. Four documents now need to exist in writing, in a form a supervisor could read cold and follow.

The parent undertaking determination, with the figures for every candidate entity and the reasoning on each of the three tests, prepared before notification rather than during the twenty-eight days. The materiality reasoning for the group-wide risk assessment, including what was left out and why. The choice of measures where a third country’s law gets in the way, with the alternatives considered and rejected. And the proportionality argument for each requirement where one is relied on, because there is no general clause to cite.

None of that is new work in the sense of new activity. It is the written record of decisions the firm was going to make anyway. The difference between the firm that has it and the firm that does not becomes visible at an inspection, which is precisely when it is too late to write it. The figures and the deadlines sit in the implementation handbook for anyone who would rather have them in one place than scattered through a standard.

The industry asked for room to exercise judgement and was given it. Judgement is the expensive option and usually worth paying for, but it should be bought deliberately rather than discovered in the third year of a supervisory cycle.

What is being measured

Article 40(2) of the sixth Anti-Money Laundering Directive requires AMLA to produce a common way of assessing the risk profile of obliged entities. Two halves: inherent risk, the risk that comes with the business being done, and residual risk, what is left once controls are taken into account. Supervisors already do something of this kind. The point of a common standard is not that somebody starts measuring, but that every supervisor in the Union measures the same things in the same way.

The financial sector received its version first, in a final report of December 2025. The standard is drafted to apply from 31 December 2027, once the Commission adopts it. The non-financial proposal went out for consultation on 13 July 2026 and closed on 27 September, with a public hearing on 10 September.

The definition worth writing down

A small entity is one with fewer than five full-time equivalent employees and an annual turnover below EUR 600 000. Both limbs, not one of them. Firms under those thresholds give a reduced set of information, aimed at what a supervisor needs in order to spot the higher-risk cases and nothing more. By default they are not asked to report on the quality of their anti-money laundering controls, although a supervisor can require the full data set for a sector it considers higher risk.

That has a consequence. Residual risk is the half of the model that depends on how good the controls are, and for the smallest firms that half will not come from the firm. Take a two-partner law firm turning over EUR 450 000. It sits under both limbs, reports the reduced set and says nothing about its controls. Its residual risk is therefore assembled from whatever its supervisor happens to hold on it. The supervisor may adjust the assessment where it holds relevant information about measures already in place, but by default it does not ask for them.

That is not entirely a gift. A small practice that has put real work into its controls has that work count only if the supervisor has seen it. That is an argument for answering a supervisor’s questions fully when they are put.

One set of questions, wherever you are

Above the thresholds the methodology sets separate data points for each category of business, AMLA’s own examples being a small legal practice, an estate agency and a football club. A notary is not an estate agent and should not be answering the same questions. The gain for anyone working in more than one Member State is that the same data points apply everywhere. A practice with offices in four countries is measured today on four national methodologies and asked for its numbers four different ways.

The date is further off than expected

For most of the sector the methodology would apply from December 2028, with the first assessments made in 2029 on the previous year’s data. For football clubs and agents it would be later still. Until then supervisors carry on with the national methodologies they use today.

Set that against the Regulation, which applies on 10 July 2027. The common way of measuring risk arrives about seventeen months after the rules themselves. AMLA’s Chair told the European Parliament in July 2026 that convergence across the Union is low, particularly in the non-financial sector, where she said meaningful results will take time.

Three things that follow

Obligations and risk scores run on different clocks. The Regulation applies in July 2027 whatever happens to this methodology, and nothing in a 2028 date postpones a 2027 obligation.

The first European assessment would be made on data from 2028, so it would find whatever the controls look like at the end of that year. There is no way to add operating history to a control built late. It is the same pattern as the selection for direct supervision, where eligibility was first counted on data as at 31 December 2025 and the final assessment runs on data as at 31 December 2026.

This is not the last word. The consultation was organised by sector, and AMLA expressly asked for proposals to reduce the burden on small entities and on national authorities. More rounds will follow, and a trade body carries far more weight in them than an individual firm. On the enforcement standards the non-financial sector barely responded the first time. AMLA ran a second consultation to reach it, and the text was amended as a result, as set out elsewhere on this page.

For anyone advising small practices, the EUR 600 000 and five-employee thresholds are the single most useful thing in the document. Not asking the smallest firms about their controls eases the burden on them. Does it also reflect how much information supervisors currently hold on this sector? Either way, the design relies on what supervisors already hold rather than on what the smallest firms report.

What they are

Article 26(5) of the Anti-Money Laundering Regulation requires AMLA to issue guidelines on ongoing monitoring of a business relationship and on the monitoring of transactions. That takes in keeping customer documents, data and information up to date. AMLA calls them one of its super priorities.

The consultation ran to 3 September 2026, and the public hearing in July drew more than 1 200 people from banks, from the non-financial sector and from civil society. AMLA said openly that it could not answer every question put to it that day. The guidelines are horizontal, covering all sectors and every size of firm, with proportionality built into the text rather than bolted on afterwards. They remain a draft, and AMLA’s timetable puts publication in the fourth quarter of 2026, so the final text may differ.

Guidelines are a different animal

A technical standard goes to the European Commission, which adopts it, and it appears in the Official Journal as law. A guideline does not. AMLA adopts it itself, it cannot change the Regulation, and it is not binding.

What it has instead is Article 54 of the AMLA Regulation. Every supervisory authority, supervisor and financial intelligence unit must confirm, within two months of a guideline being issued, whether it complies or intends to comply. If it does not, it must tell AMLA why. AMLA then publishes the fact of non-compliance, and may publish the reasons as well, giving the authority advance notice before it does. The same article requires AMLA to consult publicly on a guideline where that is appropriate, and to publish its reasons where it does not, which is why this consultation happened at all.

So the line in AMLA’s slides about national authorities indicating compliance in the first quarter of 2027 refers to that two-month clock. What it produces is a public list of which supervisors are not following which guidelines. For a firm wanting to know what its own supervisor expects, that list will say more than any speech.

One further provision in the same article is easy to miss: where a guideline requires it, obliged entities must report, in a clear and detailed way, whether they comply. A guideline can create a reporting duty of its own. Under Article 54(5), these guidelines will also replace the European Banking Authority guidelines on the same subject. The EBA guidelines stay in force until the new ones start to apply, and AMLA is required to allow a suitable transition period.

Keeping what you know current

How deep a review goes, how intense it is and how often it happens all follow the customer’s risk profile. The firm must assess how reliable and independent its sources are. A reduced depth is allowed for low-risk relationships that are not continuous or have gone inactive.

Re-collecting an expired identity document is a matter of risk, not a routine. There is no automatic re-collection. If the final text keeps this position, a firm that refreshes passports on a calendar will have a basis for reviewing that practice against risk.

Event-driven reviews are required where something risk-relevant changes: a change of ownership, a change in the customer profile, a new politically exposed person status, activity deviating from what was expected. The firm is expected to find those triggers through a monitoring framework that works, assess what they do to risk, update the information without undue delay, and reassess the risk profile where warranted.

Where a customer will not answer, temporary suspension or restriction is available on certain criteria, and termination is required if the information cannot ultimately be obtained. Alongside that sits a sentence worth pinning up: a lack of transactions does not remove money laundering or terrorist financing risk. A dormant relationship is not a solved one.

Noticing what does not fit

Use the data and the controls already held across products and services, so that nothing falls into a gap between two systems. Detect material deviations, patterns and linkages, including visible signs of sanctions evasion and not only of money laundering. Where little transaction data is held, adapt the method rather than concluding there is nothing to do. And where agents, intermediaries, group entities or networks assist with monitoring, the accountability stays with the firm.

Manual, automated and semi-automated monitoring are all acceptable. What matters is that the method fits the risk, the business model, the volume, the speed and the complexity. Manual controls must be structured and effective. Automated tools must be defined, tested and calibrated. Then the line that gives this piece its title: default settings are not a risk-based design. The draft says default settings should not be used without a documented assessment of their appropriateness. So thresholds the vendor shipped are not a risk decision until somebody has made one. The question that follows is who set them, and on what basis.

Monitoring is to be calibrated to a baseline for that customer, and customer due diligence and monitoring are to work as one thing rather than two. Peer or reference groups can support the work but must not replace understanding the customer in front of you. Material deviations should send the reviewer back to the customer information and the risk profile.

On technology, and on artificial intelligence

Technology is encouraged where it improves effectiveness. But effectiveness is to be judged by monitoring outcomes, and not by a tool’s label, its sophistication, or whether it uses artificial intelligence. Manual monitoring stays acceptable where it remains effective at the firm’s scale and risk. Outputs must be explainable enough to be reviewed and challenged, and explainability is to be sufficient for review, challenge and supervisory assessment, without requiring full technical interpretability of the model.

That is a workable standard and it cuts both ways. A firm is not being asked to open the box completely, only to say why an output happened, well enough that somebody can argue with it. It is a defence against being asked the impossible, and it removes the excuse that nothing can be explained at all.

Testing your own framework

On internal controls the draft asks a firm to test regularly against the risks in its own business-wide risk assessment, and against national or Union priorities where relevant. To know what the monitoring is meant to achieve. To keep the data accurate, validated and protected. To validate material changes to tools, rules or models. To assign ownership and fix deficiencies without undue delay.

Two further sentences do a lot of work. Focus on meaningful outputs, not alert volumes alone. A weak legacy system is not the benchmark for effectiveness. Between them they dispose of the two most common defences in this field: that the alert count proves diligence, and that the system is as good as it has always been.

What to do before they land

Sort the periodic reviews into two piles, those driven by the customer’s risk and those driven by the calendar. The second pile is where the savings are, and there is now a reason for moving them.

Find out whether the monitoring thresholds were ever set deliberately, by whom, and on what basis. If the answer is that they came with the system, write that down, because somebody will ask.

And when the final guidelines are published, read the background and rationale section rather than only the rules. AMLA asked people to do that during the consultation, and it is where the reasoning sits.

These guidelines bind nobody outside the Union, and are not binding inside it either in the strict sense. They are still the clearest written statement of what a supervisor expects of a monitoring framework. A group with a European entity will see that entity judged against them. A group without one still has a better benchmark than most internal policies.

Nobody will ask how sophisticated the monitoring is. They will ask what it found.

What the exercise is for

From 2028 AMLA will directly supervise up to forty high-risk cross-border credit and financial institutions or groups. The route to that list runs through eligibility. Under Article 12 of the Regulation that established AMLA, an entity or group is eligible if it operates in at least six Member States, including its home one. It can operate there through establishments, or by providing services across borders above certain thresholds.

Those thresholds are the arithmetic for cross-border services. In a given Member State, more than 20 000 resident customers, or more than EUR 50 000 000 of incoming and outgoing transactions generated by those customers. Either one is enough. Six Member States, counting those with an establishment and those above the thresholds, and the firm is in the pool. Being in the pool is not being selected: the pool gets a risk assessment and the highest-risk entities in it are chosen. But nothing can be selected without first being in the pool, and the pool is pure counting.

In June 2026 AMLA ran a webinar on identifying eligible institutions and published corrected slides a week later. It is the most practical document the Authority has produced: a form, a set of filters and five worked examples.

Two dates, not one

The first count used eligibility information as at 31 December 2025. Firms completed the template, returned it to their national supervisor, and national supervisors were due to send the data to AMLA by 15 August 2026. That exercise produces the list of provisionally eligible entities.

Then it happens again. By 31 March 2027, national authorities collect two further things from the entities in that provisional pool. One is the detailed data points for the risk assessment. The other is eligibility information as at 31 December 2026, which is what the final assessment runs on. So 2025 decides who enters the pool and 2026 confirms it. The first form has gone in. The second round is the one still ahead.

What counts as a customer

In a Member State where the firm holds a passport to provide services, it counts all its customers resident there, whatever product or business line they sit in. Non-active customers count. Holders of dormant accounts count. A customer holding several products with the same entity counts once, not several times. For a joint account, each holder is counted as a separate customer. For customers that are legal entities, residence means the country of establishment.

The definition proper is set out in frequently asked questions updated on 4 August. A customer is someone who maintains, or has maintained during the reference year, a business relationship with the firm. A customer is also someone who carried out at least one occasional transaction with or through it, even where there was never any relationship at all. Two consequences follow. A customer who left in March 2025 is still a customer for this purpose. And a person who walked in once, did a single transaction and never came back is a customer too.

Two sectors get their own rule. For a life insurer or a life insurance intermediary, the customer is the policy holder. For an asset management company, the customer is the person named in the shares or units register of the fund for which it is the designated manager. A holder with several lines in that fund counts once per fund.

A transaction, for the other threshold, is any operation generated by a customer that moves funds, crypto-assets, securities or other assets or value. It counts as customer-initiated where the customer gave an action or a mandate, and AMLA says that expressly includes instructions producing automated or recurring execution: standing orders, direct debits, preauthorised securities orders. One mandate, given once, generates a counted transaction every time it runs. For a book heavy in standing orders, that is the difference between clearing fifty million and not.

AMLA also says the word is to be read broadly and is not limited to the party on the other side of the contract. It covers anyone for whom the firm establishes or maintains a relationship, provides a service, executes or arranges a transaction, or acts on instructions. A transaction the firm is instructed to carry out counts as customer-initiated even where the customer is not a party to the transfer itself. In the secondary loan market that may mean the seller, the purchaser or both, and the underlying borrower as well where there is a relationship with the borrower.

Set that against how a firm’s own systems might answer the question. Dormant accounts may be excluded from management reporting. So may closed relationships. Joint accounts may be counted as one. Take a firm with 14 000 active single-name accounts in one Member State, 4 000 dormant ones, 1 500 relationships closed during the year and 900 joint accounts. On a commercial headcount it sits comfortably under the threshold. On this definition it is over it. Put the three exclusions together and a firm can cross the 20 000 line without anything changing in its business. That is the difference between a commercial headcount and a regulatory one.

Who fills in the form

Every entity that receives the template must complete and return it, including one that is plainly not eligible. For such a firm most of the sheets do not apply, and the return is how it tells its supervisor that it is not eligible. Silence is not an option, and nor is a note saying this does not concern us.

In a group, one entity reports for the whole group across the Union in a single template, listing every establishment at every level of ownership, indirectly owned subsidiaries included. Normally that is the parent undertaking, and if both a parent and a subsidiary send something in, AMLA looks only at the parent’s. Where several entities in a group are asked to report, the group appoints what AMLA calls a Designated Reporting Obliged Entity, and everyone else files only the two identification sheets naming it.

One field in the template is phrased in the negative. It asks whether another entity in the group has been designated to report, so the entity reporting for the whole group answers no. Answered the other way round, it tells the supervisor the opposite of the truth about who is reporting.

One subtlety will catch some groups. Until the Regulation applies on 10 July 2027, a holding company is not an obliged entity, so it cannot be the parent undertaking for reporting purposes. Where the ultimate parent of a group is a holding company, the reporting falls instead to the obliged entity at the highest level of consolidation.

Three traps in the counting

The first is the inactive passport. A firm holding a passport in a Member State with no resident customers there, and therefore no transactions, need not count that Member State at all. But a firm with customers there that is simply below the thresholds must still count that Member State and still report the data. Below-threshold activity gets reported even though it creates no eligibility on its own. The two situations look similar from a distance and are treated quite differently.

The second is worth checking before anything else. Incoming and outgoing transactions do not mean transactions crossing the border of that Member State, and AMLA says expressly that they should not be read that way. They mean funds or assets coming into, or going out of, the accounts of the customers reported there. The EUR 50 000 000 test therefore runs on all of those customers’ flows, not only the cross-border ones. A cross-border figure is the wrong figure, and it is wrong in the direction that understates it.

The frequently asked questions raise a question of their own. Question 33 describes the thresholds as 20 000 resident customers or EUR 20 000 000 of incoming and outgoing transactions. Question 34, the very next answer and on the same point, uses EUR 50 000 000. Did the twenty slip through? If this page understands correctly, Article 1(1)(b) of the standard sets fifty, and so does the interpretative note. A count worked from Question 33 would have tested the book against a lower line, and could have reported a Member State as in scope when it was not.

The third is a small mercy. Where a firm already has more than 20 000 customers in a Member State, the threshold is met and no transaction data is required for it.

What to do now

The first template has already gone in. The confirmation round on 2026 data is still ahead, and the full package of materials is published. It includes a video recording and five pre-filled example templates. They cover a solo entity, a group with an EU parent, a group whose parent is not yet an obliged entity, a group with several reporting entities, and a group with a head office outside the Union.

So, three things. Find out who in the group returned the first template and what number they reported. Recount it on the definition above, with dormant accounts, relationships that ended during 2025, one-off occasional transactions and joint account holders all included. Then check whether the transaction figure was computed on cross-border flows rather than on everything those customers did. And where the group’s ultimate parent is a holding company, settle now who reports in the next round.

For a group with a head office outside the Union, that head office is reported on a sheet of its own, alongside the European holding company. So somebody in the group will have to fill in those details. The decision about who reports for the group, and on what definition of a customer, is a group decision. The consequences land on the European entities, while the systems that produce the data often sit elsewhere.

None of this makes a firm more or less risky. It decides which list it is on while the question of risk is still being asked. The form has already been filed, and the only question worth asking is whether anyone checked the definition first.