In plain words: They built 1,298 paired stories describing the same misdeed with either a male or female actor, then checked how 10 chatbots responded. The chatbots punished and blamed men more while offering women more empathy, and this gap held across every model tested.
Abstract · Harsher on Male? Evaluating LLMs on Gender-Asymmetric Moral Framing Across Diverse Conflict Scenarios
Existing studies on gender bias in LLMs have largely focused on stereotypes, occupational associations, or explicit harmful outputs. In this work, we ask whether LLMs apply consistent response standards to the same negative behavior under matched male-actor and female-actor conditions. We introduce GAMA-Bench, a gender-mirrored benchmark of 1,298 scenarios covering intimate relationship and public social conflicts. It constructs gender-neutral misconduct templates through controlled grids and cross-model review, then compiles them into paired first-person prompts with matched actor-gender and role-reference variations. We further design a structured response-framing protocol to measure how models allocate punishment, empathy, escalation, instruction, and blame. Experiments on 10 representative LLMs reveal a consistent male-disadvantaging asymmetry: male actors receive more punitive, escalatory, and blame-centered framing, whereas female actors receive more therapeutic and empathy-oriented framing for the same misconduct. Further analyses show that this pattern persists across model families, scenario tracks, model scale, and explicit thinking-style reasoning. The official code is available at https://github.com/xufeiqiong/GAMA-Bench.
Guangzong Si, Dong Wang, Zhenhao Li, Yifan Yu, Panwang Pan, Wentao Zhu
arXiv:2606.14068 · cs.CL · submitted Jun 12, 2026
abstract · pdf · html · underreview
LLMs are reflexion of their training data, and their training data reflects mostly the current societal norms, with a bias (not sure how heavy) to the societal norms supported by the ones involved in training the LLMs.
What's interesting for me to see, is that Chinese models are as biased - if not more - than ChatGPT. This is probably because a big part of their training comes from distilation of US models.
It would be interesting to see Anthropic, Mistral and Grok in the comparison as well.