The Actuarial Association of Europe (AAE) has signalled its readiness to support the effective implementation of the European Union's (EU) Artificial Intelligence (AI) Act.

In its response to the European Commission on the draft implementing regulation under the EU AI Act, the association noted that actuaries, and the insurers and pension funds they serve, are among the primary downstream users of general-purpose AI (GPAI) models.

“Actuaries are extensively involved in insurance, pensions and financial risk management, where GPAI models are increasingly integrated into pricing, underwriting, reserving and operational risk processes,” it stated.

Therefore, the AAE stressed that the reliability of GPAI models is of “paramount importance” to downstream users.

“Actuaries, insurers and pension providers bear full professional responsibility for their decisions and outputs, regardless of the AI tools they employ.

"Meaningful scrutiny of GPAI models – including through this regulation – therefore serves an important public interest, and the AAE strongly supports its overarching objectives,” the AAE noted.

In its recommendations, the AAE said it would welcome further guidance on incorporating explainable AI (XAI) tools into evaluation practice. This is because it believes that XAI methodologies could help achieve a nuanced balance between transparency and business confidentiality.

Furthermore, it highlighted Article 2 of the Act, which provides for a broad range of technical access modalities, with no explicit upper bound.

On this, the association suggested it could be helpful to clarify that the means requested will be determined in a manner that is necessary and proportionate to the evaluation objective.

Although the AEE strongly supports the provisions on independent experts (Articles 3 and 4), which address independence and confidentiality requirements in detail, it does not elaborate on the technical competence criteria for expert selection.

“We suggest it might be helpful to address this gap through guidance accompanying the regulation, ensuring that appointed experts include individuals with strong statistical expertise, familiarity with uncertainty quantification and model validation, and knowledge of XAI methods — thereby enabling structured assessment without necessitating full disclosure of proprietary information," it stated.

---