Curated Insight: IAIS Application Paper on the supervision of AI

This featured resource is a paper on AI, published by the International Association of Insurance Supervisors (IAIS).

It is primarily aimed at supervisors, but naturally of interest to insurers, and some of the language is direct (e.g., “insurers and supervisors should …”).

We noted in a previous article on training and awareness that this paper was in draft; the consultation process is now complete and the IAIS issued the final paper in July 2025.

Highlights

Here are a few key takeaways from the paper:

• Oversight (3.4, page 21): human oversight is a thread through the paper, with 3.4 detailing board and senior management responsibilities.
• Proportionality (2.2, page 18): consider the actual risk level of each AI system, rather than applying blanket governance and risk management measures. For instance, a simple document retrieval AI system may not face the same scrutiny as an AI determining claim payouts. Expectations are higher for AI systems affecting retail customers than those affecting commercial customers.
• Definitions (1.2, page 10): the paper references the OECD definition of AI, and also suggests that insurers may want their own working definition. An alternate or modified definition could help create clarity and provide support in determining proportionality.
• Benefits and risks (Box 1, page 8): the paper lays out 5 key benefits (for insurers and consumers), along with 8 key risks (e.g., bias, cyber).
• Criteria to assess risks (Table 3, page 16): examples of both outcomes-related criteria (for insurers and policyholders), and model-related criteria.
• Fairness (6, page 30): guidance about preventing bias through fairness-by-design, careful data management, checking for proxies and reviewing outcomes. We have related articles on postcode proxies and on correlation and causality.
• 3^rd-parties (3.5, page 23): covers oversight of external models and concentration risk. We have related articles on 3^rd party assurance and gen AI supplier risk.

The full paper is available on the IAIS website (https://www.iais.org/2025/07/the-iais-publishes-application-paper-on-the-supervision-of-artificial-intelligence/).

Disclaimer: The information in this article does not constitute legal advice. It may not be relevant to your circumstances. It was written for specific algorithmic contexts within banks and insurance companies, may not apply to other contexts, and may not be relevant to other types of organisations.