Insights ¦ Engagement Paper Proposal for AI Live Testing

Key Take Aways

1. The FCA aims to become a smarter regulator by leveraging technology and fostering responsible AI adoption in UK financial services.
2. Introduction of AI Live Testing within the FCA’s existing AI Lab to support safe, responsible deployment of AI models in live markets.
3. The regulatory approach is principles-based and outcomes-focused, relying on existing frameworks to support innovation without additional regulation.
4. The initiative seeks to build confidence in AI model performance, enabling firms to test AI strategies and achieve positive outcomes for consumers and markets.
5. Collaboration with industry is a core element, focusing on evidence-gathering to address critical questions around model robustness, bias, explainability, and consumer impact.
6. Eligibility for AI Live Testing is open to firms with demonstrable effective pre-deployment testing, a clear roadmap for deployment, and plans for post-deployment monitoring.
7. The testing is exploratory and aims to establish shared understanding of evaluation methods, benchmarks, and mitigations across AI lifecycle stages.
8. The programme is designed to run for approximately 12 months, with each cohort testing for around six months, subject to evaluation and potential expansion.
9. The FCA emphasises a flexible, forward-looking approach to AI regulation, aligning international trends and emerging academic research on AI auditing and risk management.
10. Insights from AI Live Testing will be publicly shared (subject to confidentiality) to inform broader market understanding and support industry standards.
11. The process includes opportunities for firms to receive regulatory support such as guidance, waivers, or modifications, underlining a collaborative regulatory model.
12. The initiative aims to facilitate evidence-based assessments of AI’s societal and market impact, particularly regarding consumer protection, fairness, and model robustness.

Key Statistics

• 75% of firms in the joint BoE-FCA AI Survey are already using AI, with a further 10% planning to adopt within three years.
• The FCA’s AI Live Testing is expected to operate for approximately 12 months, with periods of active testing lasting about six months per cohort.
• The IMF models suggest up to 16% potential growth in UK economic output attributable to AI-driven productivity improvements.
• AI could raise UK national income by 5% to 14% by 2050, equivalent to over £300 billion annually in today’s terms.
• The FCA’s engagement process requires feedback by 13 June 2025, with applications opening in early summer 2025.

Key Discussion Points

• Major barriers to live AI deployment include technical challenges, governance issues, regulatory concerns, and organisational readiness.
• The proposed FCA framework addresses some deployment challenges but invites industry input on focal areas such as model robustness and consumer impact.
• Critical questions include how to assess model validity, mitigate bias, ensure explainability, and monitor outcomes continuously.
• The importance of developing shared evaluation standards, benchmarks, and performance metrics to facilitate safe deployment.
• The role of human oversight, decision protocols, and intervention mechanisms in managing AI risks during live deployment.
• The need for transparency around model documentation, decision-making processes, and audit trails to enhance trust and compliance.
• The significance of addressing potential biases, especially in sensitive areas like credit provisioning and consumer fairness.
• The value of continuous post-deployment monitoring to ensure AI models adapt effectively to changing market and consumer conditions.
• The potential for insights garnered through AI Live Testing to inform wider regulatory policy and industry best practices.
• The application of a collaborative approach, with regulatory support through guidance, waivers, or adjustments, designed to foster responsible innovation.
• Recognition that international regulators, such as NIST and Singapore’s IMDA, are exploring similar testing frameworks, although the FCA’s approach remains bespoke to the UK financial sector.
• The emphasis on constructing an evidence base and establishing technical understanding to support the safe, responsible adoption of AI in UK financial markets.

Document Description

This article outlines the FCA’s proposal for AI Live Testing, a pioneering initiative designed to enable financial services firms to trial advanced AI models in live market environments responsibly. It details the strategic importance of AI for innovation and economic growth, the principles guiding the regulatory approach, and the practical framework for participation. The document addresses eligibility criteria, testing scope, key questions for industry engagement, and how insights will be shared. It also provides illustrative examples, particularly risk management in loan and credit provisioning, highlighting how rigorous testing and monitoring can mitigate risks while unlocking AI’s benefits for consumers and markets. Overall, this article presents the FCA’s forward-looking, collaborative approach to fostering safe AI innovation within the UK’s financial services ecosystem.