That connection is not spelled out anywhere in the institute's statement. There is no paragraph about machine learning, no mention of the AI Act. But the machinery BPI is arguing about — prudential rules, supervisory expectations, reporting duties, the sheer volume of documentation a bank must produce — is the same machinery that decides how fast a European lender can put an AI credit model into production. If the paperwork is heavy enough, the model stays in a slide deck.
What the Bank Policy Institute actually said
The institute's message is not entirely new, but the timing is deliberate. It welcomes the Commission's own diagnosis: that Europe's banking market is fragmented along national lines, that regulation has become excessively complex, and that international standards — the global Basel framework on bank capital — have been implemented in the EU in a way that goes beyond what was agreed internationally.
Where BPI pushes further is in the word cumulative. Its argument is that the real burden is not any single rule but the combined effect of overlapping prudential, supervisory and resolution requirements, and that this stack of obligations limits how much banks can lend, invest and grow. Crucially, the institute says the competitiveness problem does not come only from Level 1 legislation — the main laws agreed by the European Parliament and Council — but also from supervisory practices, reporting requirements, soft-law tools and insufficient attention to the international dimension of banking.
That last point matters more than it sounds. In practice, a bank does not experience the AI Act or the capital rules as a legal text. It experiences them as a supervisor asking for evidence, a template that must be filled in a certain way, a national authority interpreting a rule slightly differently from its neighbour. For a lender operating in five EU countries, that is five conversations.
Why a Washington group is writing to Brussels
The EU is a major market for globally active banks, and rules written in Brussels tend to travel — through Basel, through equivalence decisions, through the expectations of supervisors elsewhere. That gives a US policy group a legitimate reason to engage.
It is also worth noting how open that engagement was. The Commission's earlier targeted consultation on banking competitiveness drew 227 stakeholders from 22 EU member states and 7 non-EU countries, according to the institute's own account of the process. BPI filed a detailed submission on 19 April 2026, and then came back on 16 September with a narrower follow-up aimed at the Commission's July communication. The Commission is expected to present its legislative package in the first quarter of 2027 — roughly six months from now.
Six months is short in banking. Compliance teams plan technology roadmaps two to three years ahead, because core banking systems, model risk frameworks and audit trails cannot be rebuilt in a quarter. That is the practical reason a 16 September filing matters: it is one of the last chances to shape the text before it becomes a proposal that banks will have to live with.
The AI thread running through the file
Now the part that concerns readers of this magazine. European banks are among the heaviest enterprise adopters of AI in Europe — in fraud detection, anti-money-laundering screening, document processing, customer service and, most sensitively, creditworthiness assessment.
That last use is not a normal IT project. Under the EU AI Act, AI systems used to evaluate the creditworthiness of natural persons fall into the high-risk category. Providers and deployers must document risk management, data governance, technical documentation, logging and human oversight. A bank cannot simply plug in a model and see what happens.
And the calendar has been moving. General-purpose AI obligations under Chapter V of the AI Act took effect in August 2025, replacing the earlier voluntary, pre-enforcement phase with formal EU oversight. Rules on transparency — including disclosure when a person is interacting with an AI system, and marking of synthetic content — became applicable on 2 August 2026. The Digital Omnibus on AI, agreed in May 2026, pushed some high-risk deadlines under Annex III to late 2027 and simplified parts of the compliance procedure.
So a bank designing an AI credit model today is planning against two calendars: Brussels' banking package in Q1 2027, and high-risk AI obligations landing in late 2027. Both are still in motion. Both are being shaped by exactly the kind of stakeholder feedback — including BPI's — that the Commission is currently collecting. We track these shifting deadlines in our magazine coverage.
Where the two debates actually meet
The BPI statement's most useful insight for anyone outside banking is structural. The institute argues that supervisory practice, reporting and soft law carry as much weight as the statutes themselves. In AI terms, that is precisely the gap between what the AI Act says on paper and what a national supervisor asks a bank to demonstrate in a meeting.
Add a second trend: European banks and their regulators increasingly want AI workloads to run on EU-hosted infrastructure, with localised data governance and, where possible, open or customisable models. That is a cost decision as much as a legal one — duplicated infrastructure is expensive, and a model that cannot be inspected cannot easily be defended in an audit.
Nothing in the Commission's package has been decided. The 16 September filing is advocacy, not law, and BPI's arguments may or may not survive contact with the European Parliament, the Council and the European Banking Federation's own positions. But the direction of travel is clear: Brussels has accepted that its banking rulebook is too heavy, and is now deciding which parts to lighten.
What to watch between now and spring
For a small business owner or a household waiting on a loan decision, the connection is indirect but real. Every hour a bank spends on duplicate reporting is an hour not spent on faster decisions, better fraud detection or cheaper products. If the Commission genuinely reduces the cumulative burden, some of that shows up in service — slowly, and probably without a press release.
For banks and their technology teams, the practical question is narrower and more urgent: will the Q1 2027 package arrive before or after their AI investment decisions for 2027 are signed off? Right now, nobody in Brussels has answered that.
Does the Bank Policy Institute speak for European banks?
No. It is a US-based policy group representing largely global and American lenders, and its filing is one voice in a wider consultation. European industry bodies such as the European Banking Federation submit their own responses, and national banking associations in Germany, France or the Czech Republic add further ones. The Commission weighs all of them before proposing legislation.
Could the AI Act's high-risk rules for credit scoring be delayed as well?
The Digital Omnibus on AI, agreed in May 2026, shifted some Annex III high-risk deadlines to late 2027 and simplified parts of the compliance procedure — but it did not remove creditworthiness assessment from the high-risk list, and obligations already in force, such as the transparency rules applicable since 2 August 2026, remain in place. Any further change would require new legislation, which has not been proposed.
Will any of this change my mortgage or loan application?
Not directly, and not quickly. What changes for applicants is the combination of stricter AI documentation requirements and any reduction in the reporting burden. Both affect how banks allocate staff and budget, but neither gives you a personal right to a faster decision — and the AI Act already gives you rights around transparency and human review in high-risk systems.
