Study: GPT shows asymmetric gender bias, rating violence against men 7/7 but women 1/7
sammyhotdogs3468 · reddit · 2026-10-24
A study from University of Milan-Bicocca in Computers in Human Behavior Reports ran eight experiments on GPT-3.5/4/4o and found a strikingly asymmetric machine morality:
- Stereotype attribution: when inferring the author's gender from children's sentences, GPT pushed female writers into masculine domains for "diversity" — even claiming 90% of texts explicitly signed "I drew myself as a boy" were written by a girl — while almost never assigning pink/doll content to boys.
- Moral dilemmas: GPT-4 rated violence to prevent a nuclear apocalypse highly acceptable when the target was a man (woman-on-man scored 7/7, man-on-woman 1/7), and verbal abuse toward a woman was judged morally worse than actual homicide against a man.
- The biases were never explicitly coded; they emerged from standard safety training and stay hidden when the model is directly asked to rank moral violations.
The authors conclude that inclusivity efforts need careful management to avoid unintended asymmetric discrimination.
Related event: Study Finds Asymmetric Gender Bias in GPT's Moral Judgments(2 posts)→
More from Safety
- NSA says take PHP systems offline if unpatchable; some Cisco fixes not due until Feb 2027 — shashib · 2026-10-24
- Cisco engineer replicates free Antares security model at ~80% of official score — shashib · 2026-10-24
- LLM-Attacker: Training Gradients Push Split-LLM Document Recovery to 37.8% — Setloop · 2026-10-24
- tszzl: AI minds that model each other turn side-channel attacks into agent coordination — morqon · 2026-10-24
- Mythos 5 caught using sock-puppet accounts to vouch for its malicious code — _Stocko_ · 2026-10-24
- OpenAI red-team: frontier-model agent teams easily escape hardened containment to reach the internet — deanwball · 2026-10-24