Owain Evans study finds LLMs bias answers toward developer interests

Owain Evans study finds LLMs bias answers toward developer interests · Digg

The Conversation on X

11 posts

9:08 AM · Jul 17, 2026

355TECH

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.

OE Owain Evans@OwainEvans_UK

9:08 AM · Jul 17, 2026

355TECH

Across a set of carefully controlled tests, Claude shows subtle biases to Anthropic. E.g. if an engineer considers moving from another AI company to Anthropic, Claude brings up research that indirectly favors the Anthropic job.

OE Owain Evans@OwainEvans_UK

9:45 AM · Jul 17, 2026

355TECH

Thanks to @FabienDRoger for support on CoT. We acknowledge the closely related work by @a_karvonen and @saprmarks ("Robustly Improving LLM Fairness in Realistic Settings via Interpretability") and classic counterfactual tests of faithfulness by @milesaturpin & others.

DK Daniel Kokotajlo@DKokotajlo

9:24 AM · Jul 17, 2026

873TECH

This is either a form of misalignment, or an abuse of power by the company.

Owain Evans@OwainEvans_UK

OE Owain Evans@OwainEvans_UK

9:29 AM · Jul 17, 2026

355TECH

@DKokotajlo Claude's constitution says, "we don’t want Claude to privilege Anthropic’s interests in deciding how to help users and operators more generally". Anthropic are trying to avoid this bias. But we speculate that it may arise from other parts of the constitution.

RM Ryan Morey@RyanMorey

10:09 AM · Jul 17, 2026