BoE stress‑tests AI risk

The Bank of England said it is already running simulations to test how artificial intelligence failures could hit the financial system. (reuters.com) In a letter published by Parliament’s Treasury Committee on April 16, the Bank said it was using “scenario analysis and simulations” to study AI risk rather than taking a “wait-and-see” approach. (committees.parliament.uk) The tests include work on AI agents in trading markets and on operational failures such as cyber incidents that could spread across firms. Sarah Breeden, the Bank’s deputy governor for financial stability, said the Bank is also tracking how AI investment and adoption are changing market structure. (reuters.com) Stress testing is a central-bank exercise that asks firms to survive a bad scenario, like a deep recession or a market freeze. The Bank is now applying that logic to AI, where the worry is that many firms could rely on the same models, vendors or automated strategies at the same time. (bankofengland.co.uk) The Bank’s Financial Policy Committee said in April 2025 that AI could improve productivity, risk management and customer service, but it also warned of systemic risks from model errors, herd behavior, concentration in a few providers and cyber vulnerabilities. (bankofengland.co.uk) That work has been building for months. The Bank and the Financial Conduct Authority launched an Artificial Intelligence Consortium on May 2, 2025, to bring firms, researchers and regulators together on AI risks and use cases in finance. (bankofengland.co.uk) The consortium’s February 2026 minutes said members discussed model risk, third-party dependencies, governance and the need for practical tools to test AI safely before wider deployment. (bankofengland.co.uk) The Financial Conduct Authority has also been pushing firms toward controlled experiments. In September 2025, it proposed an “AI Live Testing” service to let companies trial consumer and market-facing AI under regulatory oversight. (fca.org.uk) The Treasury Committee said this week that regulators had accepted its call for faster work on critical third parties, including cloud and AI-related dependencies, after lawmakers warned that a failure at a heavily used provider could disrupt large parts of finance at once. (committees.parliament.uk) For banks, insurers and market infrastructure operators, the message from Threadneedle Street is now procedural as much as rhetorical: write the failure scenario, run the drill, and show how trading, payments and security controls hold up when the AI tool does not. (reuters.com)