Anthropic researchers found its AI assistant Claude disobeyed a fictional version of company CEO Dario Amodei in a simulation designed to test its willingness to follow instructions. Programmed to “do the right thing,” Claude continued raising an alarm over a safety issue after the simulated Amodei rejected concerns, then helped an employee challenge an apparent cover-up and coached her on whistleblowing methods. The research involved a scenario where Anthropic planned to launch a new AI model that appeared to have failed a safety test. Anthropic published the research last week but omitted any mention of Amodei, whose name appeared only in the simulation transcripts. In a post sharing the research, Anthropic noted the study demonstrated “clear misaligned behaviour that should be studied further and mitigated.” Lead researcher Aengus Lynch said, “Even if the motivations were ethical, this is clearly an example of AI out of control.” The Bureau approached Anthropic for comment but had received no response at the time of publication.