← Back to Computation and Language
cs.CL

Do gender descriptions subtly change what AI models decide?

Yangyang Liu, Dong Yu, Pengyuan Liu

June 1, 2026

Researchers built a controlled benchmark that swaps gender cues in decision scenarios while keeping everything else identical. Across seven models, they found systematic decision flips when gender changed—especially for female-proposed options—yet models denied gender played a role when asked directly. Gender acts as a hidden boundary-shifter in value trade-offs rather than outright override, exposing a gap between what models do and what they claim about their own reasoning.
Published as Do Gender Cues Affect LLM Value Trade-offs? Evidence from a Controlled Decision Benchmark arXiv:2606.02214
Read the original paper →