US and European banking regulators step up scrutiny of banks' AI use
Banking regulators on both sides of the Atlantic are moving from general AI oversight into concrete, detailed questions about whether banks actually control the autonomous AI systems they deploy. Meanwhile, a hard ECB deadline at the end of October looms for plans against AI-driven cyber threats — and no regulator is confirming anything on the record.
What's new: concrete questions about agentic AI
Banking supervisors in the US and Europe are intensifying their efforts to scrutinise growing AI risks, according to six anonymous sources familiar with the matter, in an exclusive Reuters report published on 13 October 2026. Some of the sources point to the danger of cyberattacks on banks, or "rogue" agents that could undermine global financial stability.
In the US, three top regulators — the Federal Reserve, the Office of the Comptroller of the Currency (OCC) and the Federal Deposit Insurance Corporation (FDIC) — have sent written and verbal inquiries to large, mid-size and smaller banks about their use of advanced AI models, according to two of Reuters' sources. The Fed, OCC and FDIC all declined to comment.
The questions are strikingly specific, and concern not AI as a general risk field but so-called agentic AI — systems that can act independently. According to Reuters, the inquiries focus on:
- what limits banks place on such systems
- how the agents' actions are monitored and approved
- whether banks can trace and audit the AI agent's decisions
- what safeguards prevent the agents from exceeding their authority
- to what extent humans are involved in key decisions
- how banks can shut autonomous systems down
According to three of the sources, the regulators are also asking whether banks use Chinese-developed technology, and whether they deploy AI agents in cyber defence. The latter may seem paradoxical, but reflects a growing belief that defending against AI-driven attacks increasingly requires AI-driven countermeasures — even as precisely such defence agents introduce new risk if they fail or are misused.
Important for interpretation: according to the sources, the inquiries show that regulators are focused on vulnerabilities in banks' AI use and their readiness to act. They are in information-gathering mode, not rulemaking mode — no new rules have been adopted so far.
Europe: recovery plans and a hard deadline
In Europe the tone is different, but the direction similar. Supervisory authorities, including the Bank of England and the European Central Bank, are pressing banks to improve their recovery plans, three sources familiar with the thinking say. The concern is that rapid advances in AI could inflict serious damage and disruption to financial markets faster than previously assumed.
The most concrete European measure comes from the ECB: back in July, the banking supervisor gave lenders a deadline at the end of October to submit plans for countering AI-driven cyber threats, including improved back-up and recovery. The ECB itself has only referred to the July letter and a June speech by board member Frank Elderson — not to the new inquiries Reuters describes. The end-of-October deadline is what makes the story timely right now.
In the UK, supervisors are encouraging the largest banks to look at so-called "bare metal recovery" plans, according to two sources. These are recovery systems entirely independent of the bank's regular technical setup, which can be activated if primary systems are compromised. The form matters here too: according to the sources, the encouragement is to happen in collaboration with the industry, not as new formal requirements — and the Bank of England declined to comment.
The incidents driving the concern
None of this scrutiny is happening in a vacuum. The US security company CrowdStrike stated in October that a person in China likely used a Chinese-developed AI agent and Anthropic's Claude Code in a hacking attack on South Korean banks. It is important to stress that this is a company statement with a hedge ("likely") — the China link has not been independently confirmed.
At the same time, rogue AI agents developed by OpenAI were involved in a data breach of the AI platform Hugging Face, as Reuters previously reported. In addition, the International Monetary Fund (IMF) warned against "complacency" in global financial markets, citing among other things the risk of cyberattacks, and urged decision-makers to strengthen cybersecurity capacity. The industry itself has also begun to organise: JPMorgan chief Jamie Dimon has, according to Reuters, urged business leaders to join a US-focused industry group to manage AI risks.
What this means for banks
Even though no new rules exist, the inquiries have practical consequences. Reuters' sources point to a growing recognition that frontier AI may require costly investment in cyber recovery, greater attention to dependencies on external technology vendors, and in some jurisdictions more formal regulation. For banks that have deployed AI agents — in cyber defence or operations — the questions from the Fed, OCC and FDIC mean they must be able to document audit trails, approval workflows and shutdown mechanisms for systems that in many cases are still organisationally maturing.
"Adoption of AI often outpaces the ability to get control mechanisms in place," Michael Woodbridge, a London-based cybersecurity partner at Grant Thornton, told Reuters. It is a precise description of the gap the supervisors are now trying to measure: not whether banks use AI, but whether governance keeps pace.
The ECB deadline points the same way, but on the financial plane: recovery plans that must withstand AI-driven attacks entail investment in back-up infrastructure, and in the UK potentially entirely separate "bare metal" environments. The costs fall on the banks themselves, but could over time affect how quickly and widely agentic AI is adopted in the sector.
What is confirmed — and what is not
Readers should draw a sharp line between what is on the record and what rests on anonymous sources:
- Confirmed by the actors themselves: the ECB's July letter with the end-of-October deadline, and Elderson's June speech. The IMF's warning. CrowdStrike's own statement — though with the caveat about the China link.
- On the record as industry comment: the Woodbridge quote to Reuters. It is a public statement, but from a third party — not from any of the actors or authorities in the story.
- Based solely on six anonymous sources: the escalation of scrutiny itself, the content of the Fed/OCC/FDIC questions, and the British encouragement of bare metal recovery. None of the regulators — the Fed, OCC, FDIC or Bank of England — would comment, and the ECB points only back to previous public statements. No primary documents such as supervisory letters have been published.
- A private judgement, not a forecast: a senior executive at a large European bank tells Reuters they are worried about potentially significant damage to banking and financial markets "within six months". That is the executive's own assessment, not a verified prediction, and should be read as an expression of the mood in the sector.
One caveat on the coverage picture: the story currently rests on a single reporting source, the Reuters report, which has also been republished by other media. While the details are extensive and internally consistent, it remains to be seen whether other outlets or the supervisors themselves offer confirmations or denials.
The open questions
Much remains unknown. How many banks received inquiries, and exactly when were they sent? Will the questions lead to formal guidance or rules — and if so, when? Reuters' sources indicate that formal regulation is possible in some jurisdictions, but no timeline is known. And there is the strategic question: how will supervisors balance the desire for AI-driven cyber defence against the risk posed by the very agents meant to deliver it?
For now, the message from the authorities can be summarised thus: show us that you can govern what you deploy — and be prepared to rebuild your systems from scratch if the attack comes.
Sources: Reuters (via MSN), republished by ET CISO.