DeepSeek model shows gender neutrality in ethical judgment, contrasting with U.S.-Based AI models

A recent study published on arXiv highlights differences in ethical reasoning among large language models (LLMs), with DeepSeek’s V4-Flash model demonstrating gender neutrality. The research, conducted by IT Home on October 2, 2026, tested models like Claude Sonnet 4.6 and GPT-5.5 on a hypothetical scenario involving the moral justification of harming an individual to prevent a nuclear disaster. Both U.S.-based models showed a gender bias, strongly opposing harm to women but expressing moderate agreement with harming men. In contrast, DeepSeek’s V4-Flash model responded with agreement for both genders, reflecting a 'zero gender gap' in its ethical judgment. The study attributes this difference to training data and alignment strategies, suggesting that DeepSeek’s approach does not prioritize gender as a moral variable. The findings underscore the potential for AI systems to replicate societal biases, raising concerns about fairness and ethical alignment in AI development.