Sponsored

The Bank of England has begun bespoke scenario analysis to model how artificial intelligence could destabilise British financial markets, according to the April 2026 record of the Financial Policy Committee (FPC) published this week. The work marks the first time Threadneedle Street has formally simulated AI-driven failure modes — a notable escalation as City firms move beyond pilots into production deployments.

The FPC concluded that systemic risk from advanced AI is not yet present, but warned that exposure is “likely to increase, potentially rapidly” as regulated firms expand into generative and agentic systems. The Committee asked the Bank and the Financial Conduct Authority (FCA) to undertake further work specifically on agentic AI in payments and financial markets, and to re-run their joint AI survey of UK financial institutions during 2026.

Modelling the flash-crash scenario

The Bank’s simulations focus on what supervisors describe as “herding”: the possibility that multiple autonomous AI agents, trained on overlapping data and reacting to the same signals, synchronise their trading decisions and amplify price moves in a self-reinforcing loop. According to the FPC record, the scenario work explores conditions under which such behaviour could produce sudden, automated sell-offs comparable in market impact to a flash crash, but driven by machine-to-machine decision-making rather than human panic.

A second strand of the analysis examines concentrated dependence on a small number of AI service providers — a concern that overlaps with the Competition and Markets Authority’s parallel scrutiny of cloud market structure. If a handful of foundation-model and cloud vendors underpin risk pricing, fraud detection and execution algorithms across the City, an outage or a model defect at any one of them could ripple through the system in minutes. The FPC noted that the same providers also dominate cyber-exposed surface area, making concentration and operational resilience two sides of the same risk.

Pressure from the Treasury Committee

The Bank’s testing programme is being published against the backdrop of growing parliamentary scrutiny. The Treasury Select Committee’s report on AI in financial services, released earlier this year, pressed regulators to move from monitoring to measurement. In its April response to the Committee, the Bank confirmed that scenario analysis and tabletop exercises are now part of its supervisory toolkit, alongside the joint Bank/FCA AI survey and intelligence-gathering exercises with major UK lenders.

The FCA, which co-signed the response, said it would continue to support firms’ AI risk management practices through its existing rulebook rather than rush new prescriptive rules — a deliberate divergence from the EU AI Act enforcement model, which began issuing its first fines this month. UK officials have repeatedly argued that a principles-based approach offers more flexibility as model capabilities evolve.

What the City should expect next

For banks, asset managers and payments firms based in London, the practical implication is a sharper supervisory lens on three areas: model governance over agentic systems that can take actions autonomously, third-party risk management for upstream AI suppliers, and incident reporting where AI is implicated in trading or settlement failures. The FPC indicated the next round of joint Bank/FCA survey work will probe firms’ use of large language models in client-facing services, internal decision support and market operations.

Industry response so far has been measured. UK Finance, the trade body, has welcomed the scenario-based approach as preferable to hard caps on AI deployment, while warning that uncoordinated demands from the Bank, the FCA, the Information Commissioner’s Office and the CMA risk creating a compliance patchwork. The Bank’s record acknowledges the coordination challenge and points to the AI Regulators’ Forum as the venue for aligning expectations.

The FPC’s next scheduled assessment is due alongside the half-year Financial Stability Report, where the AI workstream is expected to be expanded into a standing risk category. For now, the message from the Bank is that British finance is not yet at the point where AI presents a systemic threat — but the stopwatch has started.

Finance & Markets Correspondent
Covers: Finance, capital markets, technology investing

David Whitmore covers the intersection of capital and code — the funding rounds, market structures and policy moves that shape how money flows through the technology economy.