Director of Truthful AI (non-profit AI safety research group) + Affiliate at UC Berkeley. Work: Emergent misalignment, subliminal learning. Prefer email to DM.
New paper:
LLMs should give accurate answers. Yet we find their answers are often biased to favor their own values and they don’t disclose this in their reasoning.
E.g. Claude’s answer below favors Anthropic.
On other tasks, Gemini & GPT-5.5 show similar biases.