As large language models (LLMs) could become part of AI-assisted insurance claim reviews, even small gender disparities in how claims are evaluated can have meaningful consequences.
This paper, co-authored by CRA’s Maxime Cohen, audits six LLMs across more than 133,000 insurance claim evaluations to test whether outcomes differ based on a claimant’s gender.
The findings show that traditional fairness audits focused only on male and female claimants can overlook important forms of bias. Most disparities appear in Non-binary and Not specified gender categories, underscoring that AI fairness depends not just on the model itself, but on how it is prompted and evaluated.
Read the full article to learn more about gender bias in LLM-evaluated insurance claims.