# LLMs Judge Men Harsher Than Women for the Same Misconduct

Original title: Harsher on Male? Evaluating LLMs on Gender-Asymmetric Moral Framing in Conflicts

> Hacker News story 50597743 · language: en · submitted 2026-11-22
> Web version: https://zeli.app/story/50597743
> About this feed: https://zeli.app/llms.txt · Agent guide: https://zeli.app/skill.md

▲ 87 · by [rzk](<https://news.ycombinator.com/user?id=rzk>) · 💬 77
[original](<https://arxiv.org/abs/2606.14068>) · [HN discussion](<https://news.ycombinator.com/item?id=50597743>)

## Summary

Guangzong Si and colleagues built GAMA-Bench, a gender-mirrored set of 1,298 conflict scenarios drawn from intimate relationships and public social settings. Each misconduct template was compiled into paired first-person prompts with only the actor's gender swapped, then scored on punishment, empathy, escalation, instruction, and blame. Across 10 representative LLMs, male actors drew more punitive, escalatory, and blame-centered framing, while female actors got more therapeutic, empathy-driven responses. The asymmetry held across model families, scenario tracks, model scale, and explicit reasoning.

## Key quote

> Male actors receive more punitive, escalatory, and blame-centered framing, whereas female actors receive more therapeutic and empathy-oriented framing for the same misconduct.

## HN discussion (77 comments)

### [notenchange](<https://news.ycombinator.com/item?id=50598688>)

> Sincerely, did anyone that knows how LLMs are trained expected otherwise?
>
> 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.

### [zdc1](<https://news.ycombinator.com/item?id=50599118>)

> I'd be curious about how the bias looks in the context of positive situations. Is the bias that LLMs hold men more directly accountable for any situation, and thus both punish negative and praise positive outcomes more strongly? Or is this bias only observable in negative situations?
>
> To be clear: it's bad either way, since women should be seen as having the same levels of agency and accountability as men, but I am curious nonetheless...

### [knighthack](<https://news.ycombinator.com/item?id=50601021>)

> This was one of the first things I noticed when using LLMs. I don't know what training or 'alignment' is preferred, but clearly there was an anti-male, rather than a neutral alignment used.
>
> You can make all sorts of misandrist statements and men-are-trash statements, with minimal or no pushback from the LLM. But a single misogynistic statement, and the LLM would turn against you, fast.
>
> Chinese LLMs seem much less biased and more neutral/equality-driven, compared to Western ones.

### [stephbook](<https://news.ycombinator.com/item?id=50598983>)

> Wasn't there a recent article on HN regarding race, gender, age and nationality bias in LLMs? I think I read it before it was flagged.
>
> The results were similar.

### [zug_zug](<https://news.ycombinator.com/item?id=50598819>)

> I did a similar microstudy on this: https://think-twice.me/is-ai-biased-or-sexist-making-chatgpt...

---

## Cite

- Page: https://zeli.app/story/50597743
- Date: 2026-11-22
- Source: arxiv.org (https://arxiv.org/abs/2606.14068)
- Hacker News: https://news.ycombinator.com/item?id=50597743

## Other languages

- zh: https://zeli.app/zh/story/50597743 · markdown: https://zeli.app/zh/story/50597743.md
- ja: https://zeli.app/ja/story/50597743 · markdown: https://zeli.app/ja/story/50597743.md
- ko: https://zeli.app/ko/story/50597743 · markdown: https://zeli.app/ko/story/50597743.md
- de: https://zeli.app/de/story/50597743 · markdown: https://zeli.app/de/story/50597743.md
- es: https://zeli.app/es/story/50597743 · markdown: https://zeli.app/es/story/50597743.md
- ru: https://zeli.app/ru/story/50597743 · markdown: https://zeli.app/ru/story/50597743.md
