Regulators testing AI shocks

The Bank of England has started running AI shock scenarios to see how advanced models could destabilize markets and banks. (committees.parliament.uk) The move became public on April 16, when Parliament’s Treasury Committee published the Bank’s response saying it would test AI agents in trading markets and study correlated behavior, or “herding,” that can intensify selloffs. Deputy Governor Sarah Breeden also said the Bank is using scenario analysis and simulations rather than a “wait-and-see” approach. (committees.parliament.uk) (insurancejournal.com) A stress scenario is a war game for the financial system: regulators invent a bad event, then test how banks, markets and key suppliers would react if it happened fast. The Bank of England said its current work looks at AI investment, AI adoption and the way AI-driven trading could amplify market moves under pressure. (bankofengland.co.uk) (insurancejournal.com) That marks a step beyond the Bank’s April 9, 2025 financial stability paper, which said AI was already being used to automate processes and could move into core decisions such as credit and insurance underwriting. In its April 1, 2026 Financial Policy Committee record, the Bank said advanced generative and agentic AI had not yet been adopted in ways that posed systemic risk, but that risks could rise rapidly as firms expand deployment. (bankofengland.co.uk 1) (bankofengland.co.uk 2) European supervisors are tracking the same problem from a different angle: operational resilience, the ability of a bank to keep critical services running during a digital failure or attack. The European Central Bank’s supervisory priorities for 2026 to 2028 put operational resilience and information-and-communication-technology capabilities at the center of oversight. (bankingsupervision.europa.eu) The European Banking Authority has reported that AI use has climbed steadily across European Union and European Economic Area banks over the past five years. Its November 2024 risk assessment said most EU banks were already using AI methods including natural language processing and neural networks, with common uses in customer support, fraud detection and profiling. (eba.europa.eu) Europe also now has a rulebook aimed at digital breakdowns. The Digital Operational Resilience Act, known as DORA, started applying on January 17, 2025 and requires financial firms to strengthen information-and-communication-technology security and resilience against severe operational disruption. (esma.europa.eu) In Britain, lawmakers are pressing on another weak point: outside technology providers. The Treasury Committee said the government still had not committed to bringing major AI and cloud companies into the Critical Third Parties regime before the end of 2026, while Treasury minister Lucy Rigby said initial designation decisions are expected this year. (committees.parliament.uk) (insurancejournal.com) The immediate trigger for the harder line was a new burst of concern about frontier AI models and cyber risk. Reuters reported that Bank of England officials said recent advances, including Anthropic’s Mythos product, sharpened fears that powerful coding systems could find and exploit software vulnerabilities inside financial institutions. (insurancejournal.com) Regulators are now treating AI less as a future efficiency tool and more as a source of shocks that can spread across firms at once. The next test is whether these simulations stay a policy exercise or get folded into the banking system’s regular stress tests. (insurancejournal.com)