Professional Risk Framework
A practical framework for pension professionals considering the use of artificial intelligence. Covers the five core risk areas every professional should understand before adopting AI in a pensions environment.
Estimated reading time: 8–12 minutes. No registration required.
Five Questions to Ask Before Using AI
These five questions provide an immediate practical check before using any AI tool in a professional pensions context. Each is explored in full below.
The Five Risk Areas
Detailed professional guidance on each risk area, including practical steps and relevant programme module references.
Data Security and Confidentiality
"Where is the data going?"
Professional Guidance
Many AI services process inputs on external servers. Entering personal member data, scheme-specific financial information or confidential correspondence into a public AI tool may breach data protection obligations, scheme confidentiality requirements or regulatory expectations.
Before using any AI tool with pension data, professionals must confirm that the data is permitted to leave the organisation, that the tool's privacy policy does not allow inputs to be used for model training, and that appropriate data processing agreements are in place.
Where any doubt exists, the professional default should be to use anonymised or synthetic data rather than live member information. This applies across all AI use cases — from drafting communications to analysing scheme data.
Practical Steps
- —Review the data privacy terms of any AI tool before use.
- —Establish an organisational policy on what categories of data may be processed through AI systems.
- —Ensure data processing agreements are reviewed by legal or compliance before tools are deployed.
- —Use test or anonymised data during evaluation and piloting phases.
Validation and Accuracy
"How will outputs be validated?"
Professional Guidance
AI systems generate outputs based on statistical patterns in training data. They do not understand pension law, regulatory guidance, scheme rules or the professional obligations of the pensions sector. They can produce confident-sounding responses that are factually incorrect, out of date or misleading.
All AI-generated outputs used in professional activities — member communications, reports, analysis, governance papers, recommendations — must be reviewed and validated by a suitably qualified individual before use or issue.
The standard of review must be proportionate to the consequence. A draft acknowledgement letter requires a different level of scrutiny than a trustee report or member benefit statement.
Practical Steps
- —Establish a review and sign-off process for all AI-assisted outputs before they leave the organisation.
- —Define which categories of output require specialist review — for example, actuarial, legal or regulatory content.
- —Maintain records of what AI tools were used and when, to support quality assurance and audit processes.
- —Never treat AI output as authoritative without independent validation.
Accountability
"Who remains accountable?"
Professional Guidance
The use of AI does not transfer accountability. Trustees remain responsible for decisions made in their fiduciary capacity. Administrators remain responsible for the accuracy of member records and transactions. Consultants remain responsible for advice provided to clients and the professional standard of their work.
Where AI has been used to support a decision or produce an output, the professional using that tool retains full accountability for the outcome. This applies regardless of whether the AI output was subsequently reviewed or whether the professional relied on it in good faith.
Accountability cannot be delegated to a tool, a supplier or a process. It remains with the individual professional and, where applicable, the organisation.
Practical Steps
- —Document how AI has been used in any professional activity where accountability may subsequently be questioned.
- —Ensure all professionals using AI tools understand that their professional obligations remain unchanged.
- —Review professional indemnity and liability positions in light of AI use.
- —Trustees should consider how AI use in scheme administration is reported and governed at board level.
Governance Controls
"Would this satisfy governance expectations?"
Professional Guidance
Organisations using AI in pension activities should have governance arrangements in place that define how AI tools may be used, by whom, for what purposes and subject to what controls. Without governance oversight, AI use can become inconsistent, uncontrolled or non-compliant.
Individual professionals making independent decisions about what AI use is acceptable creates operational risk and potential member detriment. A governance framework provides consistency, accountability and a basis for appropriate oversight.
Regulatory expectations around AI governance in financial services are developing. Organisations that establish clear governance frameworks now are better positioned to demonstrate compliance as expectations evolve.
Practical Steps
- —Develop and implement an organisational AI use policy.
- —Define which AI tools are approved for use and in what circumstances.
- —Establish approval processes for new AI tool adoption.
- —Report AI use and governance to trustees or senior leadership on a regular basis.
- —Review governance arrangements at least annually as AI capabilities and regulatory expectations develop.
Managing Errors and Consequences
"What happens if the answer is wrong?"
Professional Guidance
AI systems make mistakes. In a pensions environment, the consequences of errors can be significant. Incorrect benefit calculations, misleading member communications, flawed governance analysis or inaccurate regulatory reporting can result in member detriment, regulatory scrutiny or reputational damage.
Professionals should plan for the possibility of AI error rather than assuming outputs will be correct. This means maintaining human review at critical stages, retaining records of AI use, and having processes in place to identify and correct errors promptly when they arise.
The obligation to correct errors and protect members applies regardless of whether the error originated with a human or an AI system.
Practical Steps
- —Identify in advance the consequences of AI error for each use case.
- —Establish clear escalation and correction procedures for AI-related errors.
- —Maintain audit trails of AI use to support error investigation.
- —Include AI error scenarios in operational risk registers and business continuity planning.
- —Ensure member-facing processes retain sufficient human oversight to detect errors before they cause detriment.
Explore the Full Programme
This framework introduces the professional risk considerations explored in depth within the AI in Pensions Practice programme. The programme develops these themes through structured learning, practical examples, governance case studies and reflective exercises across six modules.
