AI in Pensions Law – opportunities and risks

However, the growing use of AI in the pensions industry also raises significant risks. Legal questions rarely turn on a single provision in a scheme’s governing documentation or an isolated piece of legislation.  Issues will often depend on complex interactions between historic scheme documentation, trust law principles, statute and case law.  An AI tool may produce a seemingly plausible answer, but if it has not understood the scheme-specific and historical context, that answer may be incomplete or wrong.

This is the central challenge for trustees and their advisers.  AI can be a useful aid.  It is not, and should not, be treated as a substitute for legal judgement.  This article considers some of the key pitfalls associated with AI in pensions law and why human judgement remains essential.

1. Benefits of AI in the pensions industry

The benefits of AI should not be understated. Industry guidance, including from the Pensions Administration Standards Association (PASA) identifies a number of areas in which AI can assist, including predictive analytics, automated documentation, data processing and member communications[1].

In practice, AI can reduce time spent on routine administrative tasks, identify inconsistencies or gaps in member records and support fraud detection by spotting unusual patterns more quickly than traditional processes. It may also help trustees, administrators and advisers work through large volumes of information more efficiently.

AI is also being used to support member communications, for example through chatbots responding to common factual queries. Used carefully, AI may therefore free up human resource to focus on more complex work.

However, these advantages do not mean AI is a substitute for legal analysis.

2. General risks associated with AI

Many of the risks associated with AI are now well known, but they take on a particular significance in a legal and pensions context.

• Lack of transparency: AI systems can be opaque, making it difficult to understand how an answer has been reached, test the underlying reasoning or demonstrate that an appropriate decision-making process has been followed.
• Hallucinations: AI may produce answers that sound authoritative but are in fact incorrect, incomplete or entirely fabricated.
• Context insensitivity: Legal analysis often turns on small but crucial details. AI tools may miss or misunderstand these nuances.
• Data quality, data protection and confidentiality: AI outputs are only as reliable as the data and inputs provided to the model. Poor quality data can produce poor quality outputs, while the use of member data or confidential scheme documents also raises obvious data protection and confidentiality concerns.

3. Risks specific to pensions law

These general risks are magnified in pensions law because of the structure of the subject matter itself. Occupational pension schemes are often governed by a long chain of documentation stretching back decades. The answer to a current question may require the reader to track how a power was drafted originally, whether and how it was amended, how statutory requirements applied at the relevant time and whether later case law changes the analysis. In many cases, the right answer cannot be reached without understanding the full legal and historical context.

This creates real risk for employers, trustees, administrators and members who use AI as if it were a source of legal advice. A tool may produce a neat summary of the law, but if it has not correctly interpreted the specific scheme documentation or the relevant legislative provisions, the conclusion may be misleading. Worse still, the output may on the surface look plausible, and the error may not be spotted until a decision has been taken or a member communication has been issued.

a) Risks for trustees and administrators

Trustees’ fiduciary and statutory duties do not change because AI is being used.  Responsibility for decisions affecting the scheme and its members remains with the trustees. If trustees rely on AI-generated interpretations of their powers or obligations without proper scrutiny, they may expose themselves to regulatory criticism, disputes and potentially claims for breach of trust.

This is not a purely theoretical point. Even apparently simple questions in pensions law require careful scheme-specific analysis. For example, does the scheme contain an augmentation power and how is it exercised? Can a particular category of member be included in an exercise? What is the correct interpretation of a historic amendment? Is a trustee discretion fettered by the rules or by wider legal principles? Can the scheme be wound up in the manner proposed? These are not questions that can safely be answered by pattern recognition alone.

The same applies where AI is used by administrators or other service providers. Trustees should understand what AI is being used for, what its limitations are, what data sits behind it and where human review will take place. As TPR explains in its statement published on 20 May 2026, trustees ‘remain accountable for outcomes even when activities are delegated’. Material decisions still need to be taken by people who understand both the legal framework and the scheme in question.

b) Risks to employers and members

The risks of AI are not only confined to trustees and administrators. Employers may use AI as a means of understanding their obligations under scheme documentation, section 75 debt legislation, consultation requirements or transaction-related pensions issues. Members may use publicly available AI tools to assess their benefit entitlements, transfer options or to assist with raising complaints. In all these cases, there is significant scope for error.

That is because generic AI answers are particularly likely to miss the scheme-specific features that matter most. A member may be told that they have a right which does not in fact exist under their scheme’s rules. An employer may be given an over-simplified answer that ignores the documentary history or the interaction between trust law, legislation and case law. Administrators may also be tempted to rely on AI-generated wording which inadvertently creates expectations or appears to promise benefits beyond those actually provided under the scheme. The result may be complaints, disputes and avoidable cost.

Conclusion

AI undoubtedly has an important role to play in the pensions industry, particularly in relation to administration, driving efficiencies and data analysis. However, pensions law is an extremely challenging environment for AI because legal outcomes so often depend on detailed scheme documentation, historical context and nuanced legal interpretation.

The real danger is not simply that AI may get the law wrong. It is that employers, trustees, administrators and members may not realise that it has got the law wrong until after action has been taken in reliance on the answer. That is why AI should be treated as an aid, not a substitute for specialist legal advice. Where legal rights, trustee powers and member benefits are concerned, careful human analysis remains essential. AI can assist but it cannot replace the scheme-specific legal analysis and professional judgement that remain essential in a rapidly changing environment.

[1] PRESS RELEASE – PASA publishes new ‘Data for AI’ Guidance to help the industry embrace innovation responsibly – The Pensions Administration Standards Association

This article is for general information purposes only and does not constitute legal advice or a comprehensive statement of the law. Specific legal advice should always be sought in relation to individual circumstances.

Meet the team:

1. 

   • James Newcome

   • Senior Associate
   • Pensions & Employee Benefits

   Visit James's profile

View the whole team