Anthropic is rolling out Auto Mode as the default for Claude Code (Pro, Max, Team) starting August 14, allowing autonomous code generation without manual approval at each step. Testing with 1,053 users showed the system detects harmful actions 89% of the time vs. humans at only 13.6%, due to habitual approval bias (97% pass-through rate). The upgrade includes enhanced security features like prompt injection filters and customizable deny rules to mitigate data leakage risks.