Banking regulators are stepping up efforts to contain growing AI risks, with some insiders pointing to the potential for cyberattacks on banks or rogue AI agents to undermine global financial stability.
The scrutiny reflects growing recognition that advances in frontier AI may require costly investment in cyber recovery, greater attention to banks’ dependence on external technology providers and, in some jurisdictions, more formal regulatory guidance, the sources said.
Three top US regulators, including the Federal Reserve, are quizzing banks about their use of advanced AI models, two of the people said, adding to earlier scrutiny. Their questions include whether banks deploy Chinese-developed technology, use AI agents in cyber defence and can control increasingly autonomous systems, three of the people said.
In Europe, supervisors including the Bank of England and the European Central Bank (ECB) are pressing banks to improve recovery plans, concerned that rapid advances in AI could cause serious harm and disruption in financial markets sooner than previously feared, three people with knowledge of their thinking said.
A senior executive at a large European bank said they, too, were worried about potentially major damage to banking and financial markets within six months.
The debate over AI oversight has intensified amid industry warnings after recent incidents, with JPMorgan CEO Jamie Dimon urging corporate leaders to join a US-focused industry group to address AI risks.
US cybersecurity company CrowdStrike said this month that a person in China had probably used a Chinese-developed AI agent and Anthropic’s Claude Code in a hack on South Korean banks. Rogue AI agents developed by OpenAI were also involved in a breach of AI platform Hugging Face.
Warning against ‘complacency’
Anthropic researchers have warned that increasingly powerful AI systems could eventually escape human control and even pose an existential threat, prompting calls for stricter regulation.
The International Monetary Fund (IMF) warned on Tuesday against “complacency” in global financial markets, citing risks including the potential for a cyberattack. It said a major attack on a critical service or financial institution could undermine confidence, and called on policymakers to boost cybersecurity capacity.
Banks are racing to adopt agentic AI, which can carry out tasks with minimal human supervision, while also grappling with the challenges posed by new models, including IT weaknesses flagged by Anthropic’s Mythos model.
“The adoption of AI is often outpacing the ability to put the controls in place,” said Michael Woodbridge, a London-based cybersecurity partner at Grant Thornton.
The US inquiries centre on how banks use agentic AI in areas such as cybersecurity, and whether AI agents are being used to identify, prioritise and respond to threats.
Questions from the Office of the Comptroller of the Currency (OCC), the Federal Deposit Insurance Corporation (FDIC) and the Fed focus on what limits are placed on these systems, how their actions are monitored and approved, and whether banks can clearly track and audit AI agents’ decisions, two of the people said.
The regulators are also asking what safeguards exist to stop agents exceeding their authority, whether humans remain involved in key decisions and how banks can shut down autonomous systems.
The OCC, the Fed and the FDIC declined to comment.
Through their questions, regulators are probing specific issues tied to advanced AI models, including autonomous agents, foreign-developed models and new cyber risks, the three people said. The written and verbal inquiries to large, mid-sized and smaller banks show regulators are focusing on vulnerabilities arising from banks’ use of AI and how prepared they are to act, they added.
Regulators are still gathering information and are not yet being prescriptive about what lenders should do, they said.
UK supervisors are separately encouraging the largest banks to examine so-called bare-metal recovery plans, two of the people said. These systems are entirely independent of a bank’s main technology setup and can take over if primary operating systems are compromised. Building and maintaining them is expensive, the people said.
Recovery and prevention
Rather than introduce new requirements, UK authorities intend to work with the industry, they added. The Bank of England declined to comment.
In July, the ECB gave lenders until the end of October to produce plans to counter AI-enabled cyberthreats, including improved backup and recovery.
Asked about its strategy for regulating AI, the ECB referred to its July letter and to a June speech by board member Frank Elderson, who stressed the need to create a sense of urgency, not alarm, about the threats.
The need for supervisory action has become more urgent after European banks and regulators expressed frustration this year at getting only limited access to Mythos, amid concerns that US banks would get first sight of frontier models to test them.
Like their US counterparts, UK supervisors are also asking banks to use frontier AI to identify vulnerabilities and speed up patching, two people said, with one adding that lenders were doing a good job so far.
The third person said eurozone banks were also being asked to map how AI is used across their businesses and third-party providers, to understand key dependencies and make sure they can switch systems off and restart them in a controlled way. Significant gaps remained at some firms, the person said.
“To protect against attacks, rapid recovery matters as much as prevention,” said Sukand Ramachandran, MD and senior partner at Boston Consulting Group.
Without separately held clean, secure copies of critical data, “a serious attack could lock customers out of their accounts and disrupt payments across the wider economy”, Ramachandran said.
In South Africa, the Financial Sector Conduct Authority warned in December 2025 that the concentration of AI capabilities in a handful of third-party vendors means a cyberattack on one provider could cascade across the financial sector, while Absa has said its defence against frontier AI cyberthreats is more AI. — Nupur Anand, Phoebe Seers, Tommy Reggiori Wilkes and Pete Schroeder, (c) 2026 Reuters, with additional reporting (c) 2026 NewsCentral Media
