Widespread Gender and Pronoun Bias in Moral Judgments Across LLMs

A comprehensive study of six major LLM families reveals systematic biases in moral judgments based on gender pronouns and grammatical markers. The research found that AI models consistently favor non-binary subjects while penalizing male subjects in fairness assessments, raising concerns about embedded biases in AI ethical decision-making.

• →Six major LLM families (Grok, GPT, LLaMA, Gemma, DeepSeek, Mistral) show statistically significant gender and pronoun biases in moral judgments.
• →Non-binary subjects are consistently favored in fairness assessments while male subjects are systematically disfavored by AI models.
• →Third-person singular sentences are more often judged as 'fair' compared to second-person constructions across all tested models.
• →The study analyzed 14,850 semantically equivalent sentences to isolate the impact of grammatical and demographic markers on AI moral reasoning.
• →Researchers attribute these biases to distributional and alignment issues learned during AI training processes.