Supervisors test AI risks

Central banks are now testing whether artificial intelligence could become a financial-stability risk, not just a cost-cutting tool for banks. (reuters.com) The Bank of England said in a letter published by U.K. lawmakers on April 16 that it is running scenario analysis and simulations to see how AI could create system-wide problems. Deputy Governor Sarah Breeden said the work includes studying whether AI agents in markets could amplify stress through correlated, herd-like behavior. (reuters.com) A scenario analysis is a drill: supervisors build a hypothetical shock and test how firms and markets react if many institutions use the same models, data feeds or service providers at once. The Bank of England’s Financial Policy Committee said in an April 2025 paper that AI could raise risks through common exposures, model errors, cyber vulnerabilities and concentration in a small number of providers. (bankofengland.co.uk) The shift follows months of official work on AI as a stability issue rather than only a productivity tool. The Treasury Committee published the Bank’s response on April 16, and the Bank said it had waited until the Financial Policy Committee’s April 1, 2026 judgments were out before replying. (publications.parliament.uk) European supervisors are sending a similar message. European Banking Authority Chair José Manuel Campa told Reuters on April 16 that banks can withstand current shocks, but he flagged cyber risk and AI-related uncertainty as areas that still require preparation. (reuters.com) That focus lands as Europe tightens rules on operational resilience — the ability to keep critical services running during a disruption. The European Banking Authority says operational resilience covers governance, outsourcing, business continuity and risk management, and the European Union’s Digital Operational Resilience Act is now the main framework for that work. (eba.europa.eu 1) (eba.europa.eu 2) The European Banking Authority has already warned that operational risk is becoming more systemic as banks digitize and rely on interconnected technology. In its risk work, it said operational-risk capital requirements have risen above 10% of total capital requirements, a sign that technology, outsourcing and disruption risks are taking a larger place in bank supervision. (eba.europa.eu 1) (eba.europa.eu 2) The Bank of England’s own AI paper made the same point in U.K. terms: the danger is not only one bad model at one bank, but many firms making similar decisions at the same time or depending on the same outside provider. That is why supervisors are moving from speeches and principles to war-game style tests. (bankofengland.co.uk) For banks, that means AI oversight is being folded into the same bucket as cyber defenses, outsourcing controls and recovery planning. For supervisors, the next step is less about whether banks use AI — most already do — and more about whether the system can keep functioning if those tools fail together. (eba.europa.eu) (reuters.com)