Anthropic is putting Claude Code on autopilot — because humans keep missing dangerous commands. Starting August 14, new sessions for Pro, Max, and Team users will use Auto Mode by default. Claude will no longer stop to ask for confirmation on every action. Instead, a safety classifier will decide in real time whether an operation is safe to run.
The move comes after Anthropic found that human approval is nearly useless in practice. The company recruited 1,053 paid professional developers and secretly planted an obviously dangerous command in their regular coding workflow. The humans only blocked it 13.6% of the time. Auto Mode, by contrast, caught more — which is exactly why Anthropic is letting the AI take over.