Pensions regulator sets AI governance framework for trustees
The Pensions Regulator is urging trustees to treat artificial intelligence as a fundamental governance issue rather than an optional feature. Its six guiding principles for AI governance, introduced this week at the Pensions UK Annual Conference, establish the first official framework for how pension schemes should manage the technology. Cliodhna Judge, the regulator’s AI strategy leader, made it clear that trustees need not become AI specialists, but they must ensure the technology is applied responsibly to benefit members effectively.
Judge emphasized that AI represents more than just another information technology tool. It should be integrated into existing governance frameworks to enhance resilience and oversight of service providers, rather than being treated as a standalone initiative. “Models evolve, risks evolve, and governance must evolve in parallel,” she told attendees. While trustees won’t be expected to examine every line of code, they will be held accountable for how providers implement AI—even when the technology operates behind the scenes.
Begin with the problem, not the solution
The first principle is deceptively simple: begin with the problem, not the solution. Judge warned against adopting AI merely because it is accessible, noting that there is a tendency for AI to become a solution looking for a problem. She urged trustees to ask how AI can be used to make things better for members and to focus on outcomes that demonstrably improve members’ positions, rather than adopting AI for its own sake.
Ongoing supervision is another critical requirement. Unlike conventional software, AI systems do not perform consistently over time. Judge noted that initial assessments fail to reveal how a model behaves in real-world conditions with actual data and processes. Continuous governance means clear ownership, monitoring outcomes, and reviewing controls when something material or unexpected happens.
Transparency needed for AI-driven benefits
Transparency and clarity are essential components. Judge framed this as a matter of trust: if AI influences a member’s pension benefits, trustees must be able to explain the reasoning behind those outcomes. This does not require rebuilding the AI model but does demand thorough documentation, audit trails, testing of evidence, and clear escalation procedures.
Data integrity forms the bedrock of effective AI use. Judge warned that AI amplifies flaws in input data—poor-quality inputs inevitably produce unreliable outputs. She stressed that “AI can improve your data, but only good governance can make it dependable,” and that trustees must verify the data feeding AI systems before trusting the results.
Collaboration to manage AI risks
Judge highlighted the need for industry-wide collaboration to understand how members are using AI and to mitigate the risks of poor outcomes. She added that TPR’s AI plan commits the regulator to engaging with industry and regulatory partners to understand the impact of AI on the pension system and to help shape a shared response.
Law firm Sackers supported this perspective, arguing that trustees do not need entirely new AI policies. Partner Hannah Savill noted that the regulator’s existing governance regulations already provide the necessary building blocks, and trustees can add an AI layer on top of existing risk-management, data-protection, cyber-security, and third-party oversight processes. She emphasized that trustees should ensure they receive information about material changes to providers’ AI use and incorporate appropriate reporting requirements into agreements.
Judge concluded that a scheme’s readiness is not measured by whether it has adopted AI, but by “how its use is understood, how well it’s governed, and very importantly, how it’s delivering better outcomes for members.” The regulator plans to publish more detailed guidance on AI governance later this year.
