The FCA AI Live Testing Service and Supervisory Engagement
The FCA AI Live Testing Service and Supervisory Engagement
The Financial Conduct Authority (FCA) launched its AI Live Testing Service to support firms in developing and testing AI systems in a controlled environment. This service allows firms to run AI models under supervision, enabling them to identify potential issues before deployment. The testing framework is designed to align with existing UK regulatory expectations, including those under the EU AI Act.
Firms using this service benefit from direct feedback from FCA specialists, which helps ensure compliance with current rules. For example, a bank testing an AI-based credit scoring model through the Live Testing Service was able to identify bias in its algorithm before launching it to customers. The FCA provided guidance on mitigating this bias, helping the bank adjust its approach and reduce risk.
The FCA also encourages firms to engage proactively with supervisory authorities through structured dialogue. This includes participating in supervisory discussions, submitting AI impact assessments, and providing updates on AI governance frameworks. The FCA’s approach is to support firms through these processes, rather than simply enforcing rules. For instance, a fintech company working on an AI-powered investment advisory tool engaged with FCA officials early in its development. The FCA provided feedback on data governance and explainability requirements, which helped the company align its AI system with regulatory expectations. This early engagement contributed to a smoother regulatory review process.
The FCA’s supervisory approach towards AI systems is evolving to reflect the increasing complexity of these technologies. Supervisors now focus on understanding how AI models operate, including their data sources, decision-making processes, and potential impacts. Firms must be able to demonstrate that their AI systems are governed appropriately, including through effective risk management and oversight. The FCA expects firms to maintain records of AI-related decisions, including any human intervention or review. For example, a financial services firm using AI for fraud detection was asked to provide documentation on how its system was trained, how it makes decisions, and how it handles edge cases. The FCA reviewed these materials to ensure that the firm had appropriate controls in place.
The EU AI Act introduces new obligations that firms must meet, including those related to prohibited practices, AI literacy, and transparency. The FCA is aligned with these requirements and expects firms to implement measures accordingly. For example, firms must ensure that AI systems do not fall into prohibited categories such as social scoring or real-time biometric identification. The FCA also requires firms to have processes in place to promote AI literacy among staff. This includes training on AI risks, limitations, and proper usage. A credit union, for example, provided AI literacy workshops to its staff to ensure they understood how AI tools were being used in customer-facing processes.
The FCA’s supervisory engagement also involves reviewing firms’ AI governance frameworks. This includes assessing whether firms have established roles and responsibilities for AI oversight, whether they have processes for monitoring AI performance, and whether they have plans for addressing AI-related incidents. The FCA expects firms to have a clear understanding of their AI systems’ limitations and to take steps to manage these. For example, a digital banking platform reviewed its AI-based chatbot to ensure it was not providing financial advice beyond its capabilities. The FCA reviewed the platform’s approach to this limitation and provided feedback on how to improve transparency for users.
The Digital Omnibus on AI, which came into force in July 2026, introduces further obligations for firms. These include deadlines for high-risk AI systems and embedded AI products. Firms must ensure that these systems meet transparency and documentation requirements. The FCA is working with firms to ensure these obligations are met. For example, a fintech company was required to implement machine-readable marking for generative AI systems placed on the market before August 2026. The FCA provided guidance on how to meet these requirements, including through documentation and audit trails.
The FCA’s approach to AI supervision is also informed by international standards such as ISO/IEC 42001:2023. Firms that adopt this standard benefit from a framework that aligns with regulatory expectations. The FCA acknowledges that firms using this standard are taking steps towards effective AI governance. For example, a financial services firm that had adopted ISO/IEC 42001 was able to demonstrate its AI governance processes to FCA supervisors. The FCA reviewed the firm’s approach to risk management, data governance, and AI oversight, and provided feedback to support continuous improvement.
The FCA’s AI Live Testing Service and supervisory engagement play a central role in helping firms develop AI systems that are compliant, transparent, and aligned with regulatory expectations. Firms that engage proactively with these processes are better positioned to manage AI-related risks. The FCA’s focus on early dialogue, practical testing, and governance frameworks helps ensure that AI systems are developed responsibly. As AI continues to evolve, firms must stay attuned to these supervisory expectations and maintain strong governance practices. The FCA’s support through testing and engagement provides firms with practical tools to meet these challenges.