Back to AI News
arxiv

Fairness Audits Expire: LLM Bias Shifts Across Model Versions, 13-Month Study Finds

Researchers repeated a fairness audit of provider-served language models in a venture screening task three times over 13 months, totaling 248,400 API calls across five providers. The same model called under the same identifier a year later showed a significantly different bias pattern, and successor models carried the same bias as their predecessors but in different size and form. Provider documentation showed models are changed or retired at the providers' discretion, with notice periods ranging from zero to six months. The authors propose "evidence anchoring" theory and recommend auditing model classes rather than single models, version records for responses, and re-audits triggered by model changes.

AI Brief

It gives organizations and regulators concrete evidence that one-off fairness audits go stale as providers silently update or retire models, with actionable policy recommendations.

arXiv:2611.20901v1 Announce Type: cross Abstract: Organizations increasingly use large language models (LLMs) for decisions about people, such as hiring, investment and employee evaluation, and rely on them over long periods of time. Fairness audits of such systems, however, are conducted only once. We examine how the biases of LLMs change over time. In a venture screening setting, models from five providers decided whether and how much to invest in startups whose founders differed only in gender, origin and university. We repeated this audit three times over 13 months (248,400 API calls). Study 1 shows that the same model, called under the same identifier one year later, displayed a significantly different bias pattern. Study 2 shows that the successors of retired models carried the same bias as their predecessors, but in a significantly different size and form. Study 3 shows that the bias does not depend on the industry of the startup. A document analysis further shows that providers change and retire models at their discretion, with notice periods of zero to six months. Drawing on these findings, we develop a theory of evidence anchoring. We theorize that bias resides in the model class and persists, whereas audit evidence is anchored to a single model and expires when the provider decides. We derive policy advice for organizations and regulators, namely audits of model classes instead of single models, a version record for every model response, and re-audits triggered by model changes.
Be the first to rate this article.

Comments & Questions (0)

Captcha
Please be respectful — let's keep the conversation friendly.

No comments yet

Be the first to comment!