FCA publishes good and poor practice paper on AI use in UK financial services (19 Oct 2026)

Published November 6, 2026

Score

Why it matters

The Financial Conduct Authority published a paper on 19 October 2026 setting out good and poor practice in the use of AI across financial services. Drawing on its supervisory work, industry engagement and AI Lab initiatives, including AI Live Testing and the Supercharged Sandbox, the paper covers governance, accountability, risk management, consumer outcomes and operational resilience, including emerging generative and agentic AI risks. Its examples are illustrative rather than exhaustive and create no new legal obligations. The paper flags firms that treat human review as a formality, where staff accept AI recommendations without real scrutiny. It also criticises firms that rely on vendor assurances without testing them, noting that they remain accountable for outsourced AI. Chief data, information and intelligence officer Jessica Rusu set out three priorities: consumer outcomes, accountability and governance, and operational resilience.

The paper does not create new rules, and its examples are not a complete list of what the FCA will expect. The FCA has committed to sharing further examples, but the paper does not specify how supervisors will test firms against them. The FCA has said it will work with industry and other regulators on a trusted framework for agentic finance, and no timetable for that framework has been published.

For firms, the practical message is that senior managers need informed oversight and clear accountability, without needing to be AI specialists. Good practice includes maintaining a full inventory of AI use cases, covering third-party and agentic systems. Attorneys advising on AI should treat the paper as the regulator's clearest statement yet on how existing rules apply, and check client vendor contracts for testing and audit rights. The AI Live Testing evaluation report is due in Q1 2027 and is the next milestone to watch.