PNAS Study: LLMs Amplify Gender Bias Both Writing CVs and Screening Them

AnnaCiaunica · x · 2026-11-19

A PNAS audit study, "Automating inequality? Gender bias in AI-mediated labor markets," randomly varied applicant gender while holding all else constant and found LLMs generate gender-stereotyped CVs and exhibit bias in screening scores and wage recommendations. Screening tracks occupational gender segregation—penalizing women in male-dominated fields and favoring them in female-dominated ones—while wage recommendations mirror real-world gender wage gaps; explicit gender cues in prompts aren't needed, as name-based signaling suffices.

The authors identify two compounding channels: evaluation bias (identical content scored differently by gender) and generation bias (gendered model output affecting later evaluation). Because the same systems shape CVs and then assess them, these risks constitute measurable algorithmic amplification in AI-mediated job search.

Original post →

More from AGI Musings

AGI Musings channel →