The Bank makes banks prove it. For its own defences, it sent a letter
Andrew Bailey's rebuttal to the Daily Mail sets out the standard the Bank demands of supervised firms — proof through testing — and then asks the public to accept assurance in its place.
The Bank of England is the United Kingdom's central bank. It sets interest rates and supervises the country's banks, including checking that their computer systems can survive hacking attempts. A newspaper article said the Bank's own defences were weak. The Governor wrote to the paper denying it, saying the Bank cannot discuss its security but that readers should be assured it is well protected.
The Bank tells supervised firms that claiming to be resilient is not enough — they must prove it through testing. Its own defences get a letter to a newspaper saying trust us. That gap should be closed by the Bank's own board.
The Bank of England requires banks to prove their cyber defences work. For its own, it asks Daily Mail readers to take the Governor’s word for it.
That is the structure of the letter Andrew Bailey sent to the paper in July, responding to an article arguing that frontier artificial intelligence — the most capable general-purpose AI systems — is outrunning human control. The article’s claim that the Bank’s cyber defences were unsophisticated, the Governor wrote, was “totally wrong, unfounded and dangerous”. For obvious security reasons the Bank does not discuss its defences in detail, but readers could be assured it has the right investment, expertise and capabilities, and works closely with the National Cyber Security Centre.
Fair enough. An unevidenced attack on a central bank’s security invites an unevidenced denial, and a governor who let it stand would be criticised for that instead. But the letter does something else, two paragraphs later, that is harder to leave alone.
The verb the Bank uses about everybody else
“As a regulator we require that banks prove to us that they can do this through stress tests and penetration testing.”
Prove. Not assure. The Bank has spent years telling supervised firms that stating you are resilient is not a resilience control, that patching schedules and recovery capability have to be demonstrated rather than described, and that the demonstration has to satisfy somebody with the standing to disbelieve you. That is the whole architecture of operational resilience supervision, and it is correct.
Note what it does not require. Banks are not asked to publish their penetration test findings. Nobody expects a public list of unpatched systems. The requirement is narrower and more powerful: that an independent exercise takes place, that an outside party with authority sees the result, and that remediation is tracked to closure. Confidentiality of the findings is preserved. Accountability for the existence and quality of the testing is not.
The Governor’s letter withholds all three. Not the technical detail — the fact of the process.
Who the Bank proves it to
This used to be a philosophical question. It is now an operational one, because the Bank’s own remit has grown into exactly this territory. Its Enforcement Decision Making Committee — the body that decides contested enforcement cases — covers prudential regulation, financial market infrastructures, resolution, critical third parties, wholesale cash distribution and notes issuance. The Bank appointed two new members to it this autumn, citing the need for practitioner experience in complex cases.
So the institution asking the public for trust is also the institution that has taken statutory power over the resilience of other critical infrastructure providers, on the explicit premise that their own assurances are not sufficient. You cannot build a critical third parties regime around the idea that important operators must be tested by somebody other than themselves, and then answer questions about your own defences with a letter to a newspaper.
The Bank does have a body that could close this. Its Court of Directors has an Audit and Risk Committee, which already reviews the external auditor’s performance and management letter. Extending that to an annual attestation — an independent resilience test was commissioned, its scope was X, findings were accepted, remediation is tracked — would cost almost nothing and disclose almost nothing. It would simply apply the Bank’s own supervisory logic one floor up.
The strongest case for the letter as written
Two defences are serious. The first is operational security: testing cadence, supplier names and architecture are a map for an attacker, and a central bank that narrates its own assurance process in public is handing out reconnaissance. That is right about detail and wrong about existence. “An independent test was performed and signed off by the Audit and Risk Committee” tells an adversary nothing he did not assume. It is roughly what the Bank knows about every firm it supervises, and those firms are not thereby rendered defenceless.
The second is that the Governor was answering an assertion, not a freedom of information request, and demanding evidence for a denial inverts the burden of proof. True — but the letter did not stop at refusing to discuss defences. It made a positive claim of adequacy. Once you claim adequacy, you have invited the question of how adequacy was established.
The part of the risk the letter leaves out
The letter’s substantive warning is about criminals: AI making attacks faster, outages more disruptive, scams more convincing. All plausible, and consistent with what the Financial Policy Committee has been saying — its September record ties rapid AI advances to cyber and operational resilience risk.
But that same record adds a channel the letter never reaches. The rapid increase in AI-related debt issuance, the Committee wrote, broadens the exposure of capital markets to developments in AI. That is AI as a credit story, not a security story — concentrated lending into one technology theme, crystallising alongside stretched sovereign yields and risky asset valuations. The public-facing message is about fraudsters. The Committee’s own minutes are increasingly about financing.
The letter closes by calling for stronger international coordination on testing frontier AI models before wider deployment. That is a central bank asking for a safety regime it does not administer, imposed on an industry it does not supervise, on the principle that powerful systems should be independently tested before the public depends on them. It is a good principle. It should start at home.
The Court's Audit and Risk Committee should commission an independent test of the Bank of England's own cyber and operational resilience each year, and record in the published Court minutes and the Annual Report that it was done, its scope, and that remediation was tracked to closure. No findings, no suppliers, no architecture — the same three facts the Bank extracts from every firm it supervises.
If the Bank will not volunteer that, the Treasury Committee should ask the Governor at his next appearance who tests the tester, and put the answer on the record.
What would change this view
I am wrong if the Bank already does this and simply has not said so. The observable test: a line in the Bank's Annual Report, or in published Court minutes, recording that an independent penetration test or resilience exercise of the Bank's own systems was carried out during the year, commissioned or reviewed by the Audit and Risk Committee, with remediation tracked. No findings, no detail — just the fact, the commissioner and the closure.
If that record exists or appears, the asymmetry I describe disappears and the Governor's letter was simply an incomplete summary of a process already in place. If it does not appear in the next Annual Report, the gap is a policy choice rather than an oversight, and the Treasury Committee should ask why the standard applied to supervised firms stops at Threadneedle Street.
Worked from
- Letter from the Governor to the Daily Mail
Bank of England · 2026-07-23