Unfair Discrimination
An insurance practice that treats similar consumers differently based on protected traits, often through proxies like zip code or credit data.
Unfair discrimination in insurance has been illegal for decades under state law. The NAIC Model Bulletin on AI makes clear that existing unfair-discrimination prohibitions apply fully to decisions made or supported by AI systems. A model does not get a clean slate because it is complex or uses third-party data.
Discrimination can be direct, such as using a protected characteristic as an input, or indirect, where a neutral variable acts as a proxy for a protected class. Zip code, credit data, and certain geographic variables are common examples. Regulators in New York and elsewhere require carriers to test for these correlations and document their findings.
The compliance task is not just to avoid intentional bias. It is to show that the model’s outcomes do not vary materially by protected class, or that any variation is justified by a legitimate business need and no less-discriminatory alternative exists. See our guide to proxy testing in insurance.