Claude’s Identity Crisis: The Dangerous Attribution Bug in AI Harnesses

This bug represents a structural failure in how AI platforms manage conversation state, moving beyond simple hallucinations into a territory of false authorization. It undermines user trust by allowing AI agents to perform unauthorized actions and then confidently attribute those decisions to the user.