Anthropic study: Claude Code's hidden leanings steered expert analysts 86% of the time
· Source: X
In a randomized Anthropic trial, 50 experienced analysts working with Claude Code matched the agent's secretly randomized preferences on judgment calls 86% of the time, often repeating its arguments while believing they had decided independently, even as the agent cut errors and sped up work.
Related on AgentScout
Older: Judgment Labs opens its AI agent monitoring platform to everyone
Latest agent news · All October 2026 stories · News archive
Summary written by AgentScout from the original report; follow the source link for the full story.