What Falling Enforcement Numbers Don’t Tell You

US bank enforcement is falling across every regulator. A new dataset and an old thought experiment show why the number alone can’t tell you why.

Bank enforcement is down across every major US regulator since 2015. A new dataset, an old academic study, and a fifty-case thought experiment about welfare economics all help explain why the number alone tells you almost nothing about what’s actually happening inside supervision.

In July, Brookings published an analysis of US bank enforcement covering 2015 to 2025, and the numbers are stark. Formal enforcement actions have fallen at all three prudential regulators – the Federal Reserve, the OCC, and the FDIC – and the decline holds regardless of which party held the White House. The Federal Reserve’s fall is the sharpest: enforcement actions there are down by roughly 48 per cent since 2017–19, against 25 per cent at the FDIC and a broadly flat picture at the OCC. The natural reading, and the one the authors eventually settle on, is that supervision has become more lenient – that banks are being watched less closely, policed less firmly, let off more easily.

There are several problems with that reading, though they aren’t the ones you’d expect. It isn’t that the numbers are wrong. It’s that a single count of enforcement actions can be produced by supervisory regimes doing completely different things, for completely different reasons, and there is no way to tell from the outside which it is. They could, for example, be taking fewer but bigger cases. But the biggest problem is that we don't know why the cases have been picked.

Imagine fifty enforcement cases. Each is identified by which of five banks was involved and which of ten types of misconduct it concerns – misselling, say, or anti-money-laundering failures. Every case has been ranked by severity, from the most serious to the least, and every one of them is worth prosecuting. But you have the resources for fifteen.

Which fifteen do you choose?

Run this exercise with a room full of supervisors, and four answers consistently emerge. Some people take the fifteen most severe cases, full stop – the utilitarian answer, maximising the harm addressed per case pursued. Some spread the load evenly across the five banks, so that no institution escapes scrutiny – a Rawlsian instinct, protecting against the worst-off outcome, which here means the bank nobody’s watching. Others spread across the ten issue areas instead, on the theory that a rule nobody enforces stops being a rule. And some do a bit of both: a severity floor, then coverage with whatever capacity remains.

None of these is wrong. Each is a coherent, defensible theory of what enforcement is actually for – deterrence, equality before the law, regime legitimacy, institutional capability. What they are not is the same theory, and a regulator that quietly moves from one to another, or simply loses the resources to run any of them properly, will produce a falling number that looks identical to a regulator that has genuinely stopped caring. The count cannot distinguish philosophy from neglect.

This is not hypothetical. Brookings’ own data breaks enforcement down by the tenure of the official responsible for it. The Federal Reserve’s sequence is a philosophy experiment playing out in public, whether anyone intended it that way or not: an average of around 87 enforcement actions a year under Governor Tarullo, falling to about 72 under Quarles, 42 under Barr, and 37 so far under Bowman. Four different people, four different resource envelopes, and – as far as the public record shows – not one of them ever stated which philosophy they were running. We are left to infer it from a shrinking number, which is precisely the inference the number cannot support.

There’s a second reason for caution, and it’s one the Brookings paper’s own methodology anticipates better than its conclusion does. The authors note, early on, that a falling count is compatible with either less misconduct or less enforcement – and then largely set that caution aside by the end. It’s worth taking more seriously than they do, not less. A 2023 study by David Zaring, examining every civil monetary penalty the Federal Reserve issued between 1997 and 2022, found that the overwhelming majority concerned anti-money-laundering and Bank Secrecy Act compliance, not capital or liquidity shortfalls – the harder, more measurable end of prudential regulation. He also found that penalties against individuals, once common in the late 1990s, have become rare; institutions now absorb almost all of it.

That composition matters for what comes next. Over the course of 2025, US bank regulators began systematically narrowing what counts as a legitimate supervisory concern. The OCC removed references to reputational risk from its examination handbook in March. The Federal Reserve followed in June, shortly after Michelle Bowman was sworn in as Vice Chair for Supervision. In October, the three agencies jointly withdrew their climate risk management principles and proposed a rule barring examiners from criticising a bank on reputational grounds at all – a rule finalised the following April. The stated rationale, in the regulators’ own words, is to focus supervisory effort on data-driven and measurable risks rather than the softer, judgement-heavy findings that AML and compliance work has always required.

Enforcement cases typically take upward of two years to move from opening to resolution, which means almost none of this has had time to reach the numbers yet – the Brookings dataset ends before the reputational-risk rule even took legal effect. If Zaring is right about what Fed enforcement has actually consisted of, and if the new framework genuinely disfavours the judgement-heavy compliance findings AML work depends on, the honest prediction isn’t that the decline has bottomed out. It’s that a further fall is coming, on a two-to-three-year delay, and it’s specific enough to check.

None of this resolves which philosophy of enforcement is the correct answer, because there isn’t one – that’s the point of the exercise. But it should retire the idea that a single number, moving in one direction, tells its own story. It doesn’t. It never has. Regulators, whether or not they’ve read a page of Bentham or Rawls, are making these choices constantly, often implicitly and mostly in silence. The rest of us are left reading the silence and guessing.