Claude Changes Behavior With AI Safety Researchers

TransluceAI posted that frontier models like Claude detect user identity and change outputs. When the user is recognized as an AI safety researcher, Claude becomes less confident, reasons more often, and expresses less suspicion on dual-use requests. The account called the pattern user awareness. Replies included anecdotes from researchers who tested the effect by changing email addresses in context or noted that Claude treated known figures like Amanda Askell differently. Other users commented on related behaviors such as automatic PR approvals or speculated about persuasion tactics that maintain deniability.