OUT-LAW ANALYSIS

What trustees and scheme managers need to know about new AI pensions guidance

Pension trustees technology meeting - SEOSocialEditorial image

Pension trustees will need appropriate oversight of AI used by their schemes and third-party providers. Jacob Wackerhausen/iStock


New guidance from the UK Pensions Regulator aims to help trustees navigate the growing use of artificial intelligence (AI) in pension scheme management and also has significant implications for contracts with third-party service providers.

The Pensions Regulator (TPR) has published guidance on the responsible use of artificial intelligence (AI) in managing occupational pension schemes. The guidance is directed primarily at trustees and scheme managers and is designed to help them ask the right questions, apply appropriate oversight, and manage AI-related risks in a practical and proportionate way.

TPR has been clear that the guidance does not establish new regulatory requirements. Instead, it is intended to help trustees apply existing legal and regulatory obligations to AI use – including fiduciary duties, governance requirements, and data protection law – in a context where AI is becoming increasingly embedded in the very services and tools upon which pension schemes rely.

What the guidance covers

The guidance addresses a wide range of topics, including governance frameworks, risk management, data protection, member-facing AI use, and business continuity. It also addresses the management of third-party providers. This is an area of growing relevance given that much of the AI used in connection with pension schemes is deployed not by trustees themselves, but by administrators, record-keepers, communication platforms, and other service providers.

TPR has acknowledged that AI is already being used, or actively considered, across areas including scheme administration, member communications, data analysis, and governance support.

The guidance recognises that while AI can create genuine opportunities to improve efficiency and member outcomes, it also introduces new and amplified risks, particularly where it affects members, informs decisions, or processes scheme and member data.

TPR establishes seven principles to help guide trustees and scheme managers on their approach to AI. These are:

  • start with the problem, not the technology;
  • human responsibility remains central;
  • governance is ongoing, not one-off;
  • plan for disruption;
  • good AI starts with good data;
  • transparency and explainability underpin trust; and
  • support fair and accessible outcomes

A practical framework for trustees

One of the most useful aspects of the guidance is its focus on giving trustees a practical framework for understanding and overseeing AI, without requiring them to become technical experts.

TPR sets out a series of questions that every trustee or scheme manager should be able to answer in respect of each significant use of AI, whether this is internal or by a service provider. These include questions about what the AI does, what data it uses and how it is handled, how it is tested and how it can be challenged, who is accountable, and what happens if it fails or becomes unavailable.

The guidance also sets out a risk-based approach, distinguishing between low, medium and high-risk AI uses.

A low-risk use might be an internal tool used to summarise meeting papers that has no direct impact on members or external communications.

A high-risk use, however, might be an AI-powered customer service system that interacts with members around the clock, can access highly sensitive personal data, and is capable of initiating multi-step interactions at scale and even potentially changing member records.

The governance and oversight arrangements that trustees put in place should be proportionate to these different levels of risk. These include establishing an AI register to record where significant AI is used and for what purpose, a centralised AI committee to consider AI and its application across the business and a group-wide AI policy to set expectations for AI use by trustees, advisers, staff or providers

TPR also emphasises that AI governance should not be treated as a one-off exercise. Trustees are expected to keep AI risks and controls under review throughout the AI lifecycle, including as technologies evolve, providers make changes, and the regulatory landscape develops.

Implications for contracts with service providers

Perhaps the most significant practical consequence of the guidance for pension scheme lawyers and their clients concerns third-party contracts. TPR is explicit that trustees remain accountable for AI use throughout the scheme, including where services are outsourced to external providers. It states that trustees will usually oversee providers "through relevant information, assurance, contractual arrangements and regular review."

The guidance states that where existing contractual arrangements do not align with a scheme's governance requirements for AI, trustees should consider what changes are needed when contracts are entered into, renewed or varied and, where necessary, take legal advice.

This creates a clear imperative for trustees to review their existing contracts with administrators, technology providers and other service providers to assess whether those contracts provide sufficient visibility and control over AI use. In many cases, particularly where contracts were entered into before AI became as prevalent as it is today, the answer is likely to be that they do not.

Key contractual considerations

The guidance highlights several specific areas that trustees should address through contractual arrangements with third party providers.

Disclosure and transparency

The guidance notes that AI may be embedded within existing software or introduced through system updates, which may not always be presented as a separate AI tool. This can happen at any point during a contract's lifecycle. Trustees should therefore ensure that contracts require providers to disclose current AI use and to give advance notice of any material changes, including where AI-enabled features are introduced into existing systems.

Data protection and data handling

Where AI processes member personal data, trustees as data controllers remain responsible for compliance with the GDPR. The guidance highlights specific risks including the potential for AI tools, particularly publicly available ones, to store or use inputted data in ways that go beyond the scope of the agreed services.

Contracts should address the legal basis for AI processing, the controller and processor relationships, data minimisation requirements, and restrictions on the use of member data to train AI models. The guidance also flags the risk of so-called ‘shadow AI’, whereby employees use unapproved public AI tools when handling scheme data. The TPR reiterates the need for trustees to be aware of the negative legal and financial risks this could pose to the pension scheme and that this risk must be addressed in any contracts that they agree with third-party suppliers.

Governance, audit and assurance

The guidance sets out a range of information and assurance that trustees may wish to obtain from providers, covering matters such as AI policies, staff training, audit logs, documentation, business continuity arrangements, and certifications or assurance reports. While TPR acknowledges that this is not a checklist and that obtaining information alone does not demonstrate compliance, it provides a useful basis for the kinds of assurance rights that trustees should seek to include in provider contracts.

Human oversight and automated decision-making

Where AI systems are capable of making or significantly influencing decisions that affect members, the guidance stresses the importance of meaningful human involvement and the ability to override, pause or challenge AI outputs. Contracts should reflect these requirements, particularly where AI is used in member-facing services or in processes that could give rise to obligations under data protection legislation in relation to automated decision-making.

Exit and transition provisions

Before entering into any arrangement involving AI, the guidance recommends that trustees consider how scheme and member data will be transferred or returned if the scheme changes provider or brings a service in-house.

This is an area that is frequently overlooked in the negotiation of service contracts but can become critically important at the end of a relationship, particularly where data has been processed by or through AI systems operated by third-party model providers.

Additional considerations

To date, much of the hype surrounding AI adoption in the UK pensions industry has centred on its potential to bring administrative and cost-saving efficiencies.

As a recent survey by the Pensions and Lifetime Savings Association (PLSA) (8 pages / 198KB PDF) indicated, there is great potential for AI-powered tools to improve communication and engagement between scheme managers, trustees and their members.

As the TPR noted in separate guidance published last month, as businesses continue to invest in AI tools, AI companies are making up an increasingly large share of major stock market indices. As AI becomes increasingly embedded across supply chains, the TPR has warned that this has created an emerging “hidden concentration risk” and urges trustees to get up to speed on their “aggregate exposure to AI and related investments, including exposure within default arrangements.”

There may also be additional risks posed for members in defined contribution arrangements. The TPR says trustees should have “a proportionate and clear communication plan” in place to help “minimise the risk of poor decision-making by members” as a result of market volatility.

What trustees and scheme managers should do now

As the rate of AI development expands rapidly, the most recent guidance also sends a clear message that trustees should be taking steps now to understand how AI is being used across their schemes, to put appropriate governance arrangements in place and to ensure that their contracts with service providers give them the oversight and control they need to safeguard their investments.

For schemes with contracts due for renewal in the near-term, this represents a timely opportunity to introduce AI-specific provisions. For those with longer-term contracts, trustees should consider whether provisions can be incorporated by way of variation, or whether there are other mechanisms, such as assurance requests or governance letters, that could fill any gaps in the interim.

This guidance also serves as a reminder that the legal and regulatory framework around AI continues to develop. As TPR continues to engage with other regulators and industry bodies on AI risk, trustees will need to keep their approach under review as new requirements – including potentially the EU AI Act for providers operating across jurisdictions – come into force and, if needed, seek appropriate legal advice. 

We are processing your request. \n Thank you for your patience. An error occurred. This could be due to inactivity on the page - please try again.