FCA publishes AI good and poor practice and asks LLM providers to add disclaimers
FCA data chief Jessica Rusu used a Bloomberg policy speech to publish observations on good and poor AI practice from this year's firm testing, call on general-purpose AI providers to add disclaimers, and set out how the FCA will supervise agentic finance. For compliance and technology leads, the material change sits in incident reporting and accountability expectations rather than any new rulebook.
- The FCA published good and poor AI practice today, structured around consumer centricity, accountability and governance, and operational resilience.
- Rusu said existing rules requiring firms to report major incidents must extend to significant AI failures and near-misses.
- Six of the seven Mills Review recommendations are being embedded into live FCA work immediately, with the response focused on testing, assurance, monitoring and surveillance rather than further consultation.
- The FCA called on general-purpose AI providers to add clear, proportionate and consistent disclaimers about the limits of financial information and advice their models generate.
- The FCA is building AI supervisory tools, including a market first responder agent for about 9,000 small wholesale transaction-reporting firms and tooling for 60,000 AML firms entering its remit in 2028.