Claude Code Auto Mode Is Now the Default — and Outperforms Human Review

Starting August 14, Anthropic is enabling Claude Code's auto mode by default for Pro, Max, and Team subscribers: the agent completes tasks on its own without asking for approval at each step — only before irreversible or destructive actions. A test with 1,053 users produced a surprising result: auto mode blocked 89% of potentially harmful actions, while humans waved through 97% of approval requests without looking and caught only 13.6% of real threats.

Human oversight of AI agents has been officially recognized as less reliable than the model's built-in guardrails. This marks a turning point for the entire industry: safety is now built around the model's own rules rather than approval buttons — and that is how most next-generation agentic products will be designed.