Pension schemes have both an opportunity and a duty to deploy artificial intelligence (AI), but its use must be subject to strict safeguards, according to Aon Netherlands wealth director, Frank Driessen.

Speaking to European Pensions, he said the rise of generative AI adds an “entirely new layer” to pension communication.

“Members are increasingly confronted with highly assertive, seemingly authoritative answers from systems that have no knowledge of their specific pension scheme, tax position or family circumstances. This creates a new tension: members experience these outputs as ‘advice’, while in reality they are generic, unvalidated information,” he warned.

His comments come at a time when increasing numbers of pension savers across Europe are turning to AI for pension information and advice.

In March, the Belgian Pensions Ombudsman identified the use of AI by future pensioners as a "new development" in complaints received in 2025, noting that pension savers are increasingly seeking information through AI tools.

Against this backdrop, Driessen set out three safeguards for pension schemes and administrators deploying AI on their own portals and websites.

First, he said AI must operate exclusively on validated, up-to-date information relating to the specific pension scheme as well as relevant legislation and regulations.

Second, he stressed the need for full transparency about what AI tools can and cannot do, arguing that while providing explanations and insights is acceptable, offering personalised tax advice is not.

Third, Driessen said a human must remain involved in decisions involving far-reaching or irreversible choices, suggesting this could be achieved through measures such as a conversation with an adviser or a mandatory telephone confirmation.

Driessen warned that the pensions sector “cannot hide behind the argument that a member ‘simply relied on an external AI tool’”.

Indeed, he argued that the sector has a responsibility to actively intervene when it is clear that members are at risk of making poor decisions based on misleading AI information.

This means asking probing questions, explaining what actually applies within the specific scheme and, where necessary, building in additional checks or cooling-off periods.

“This requires pension advisers and front-office staff who recognise the influence of AI and systematically ask follow-up questions such as: ‘What exactly are you basing this on?’ and ‘Let’s compare this assumption with the specifics of your own scheme’,” he stated.

However, he stressed that intervening does not only mean preventing an unfavourable choice, but rather explaining more clearly, building in additional checks and, where necessary, creating reflection time.

“If complaints or customer interactions reveal that the same AI-driven misunderstandings repeatedly arise around certain topics, such as partners’ pensions, high/low benefit structures or tax thresholds, this is a clear signal to improve both communication and choice architecture,” he said.

In practice, he recommended that core information from pension schemes be made “AI-proof” not only to help people, but also the AI systems, reach the correct conclusions.

He added that AI functionalities operated by pension administrators and pension funds within their portals should be fed with validated data and operate within clearly defined boundaries, preferably with a human in the loop for irreversible decisions.

He concluded: “AI does not make the pension landscape simpler; it makes its complexity more visible and increases the risk that members overestimate their own understanding. Ignoring or banning AI is not a viable response – members will use it regardless.

“The challenge for the Dutch pension market is to make its own information provision so strong, consistent and accessible that both members and the AI systems they rely on take it as their reference point.

“That requires better communication, sharper design choices and an active dialogue with members. Those who take this responsibility seriously can turn AI from a risk into a useful tool.”

---