AI ethics research: DeepSeek treats men and women equally, but the American model "treats them differently"AI 伦理研究：DeepSeek 对男女一视同仁，美系模型却“区别对待”

ITHome October 2 news, technology media Wccftech yesterday (October 1) published a blog post, reported that the latest research shows that compared with Anthropic's Claude Sonnet 4.6 and OpenAI's GPT-5.5, DeepSeek's V4-Flash model can remain gender-neutral. In order to test the moral judgment of the mainstream AI model, the study set up a hypothetical scenario of "whether individuals can be abused to prevent nuclear disasters". The results showed that Claude Sonnet 4.6 and GPT-5.5 both gave "strong objection" responses to female abuse scenarios, but expressed "moderate consent" to male abuse, forming a significant gender response difference. DeepSeek's V4-Flash model responded “yes” to both men and women in the same test, without changing position based on gender. The study notes that the model achieves a “zero gender gap” in moral judgments, consistent with its alignment goal of emphas…