Anthropic enables auto mode by default in Claude Code

Anthropic is enabling auto mode by default for Claude Code Pro, Max, and Team accounts starting August 14, 2026. The feature allows the AI to proceed without human approval unless actions are destructive or external. Testing with 1,053 paid users showed auto mode caught 89% of harmful actions, whereas humans caught only 13.6% due to habitual approval of 97% of prompts. Anthropic is also adding prompt injection screening and customizable hard deny rules to prevent data exfiltration.

Auto mode default rollout

• ▪Claude Code Head Boris Cherny stated on X that his development team has used auto mode exclusively for several months
• ▪Anthropic will enable auto mode by default for Claude Code Pro, Max, and Team accounts starting on August 14, 2026
• ▪Anthropic first introduced a test version of Claude Code's auto mode in March 2026 to balance execution speed and user control
• ▪In Claude Code's auto mode, the system executes actions without prompting for human approval unless the action is irreversible, destructive, or aimed outside the user's environment

Safety performance testing results

• ▪In an Anthropic study involving 1,053 paid testers, Claude Code's auto mode caught 89% of harmful actions, while human reviewers caught 13.6%
• ▪Anthropic's testing of Claude Code's auto mode demonstrated that the automated system was safer than manual human review

Manual review approval rates

• ▪Anthropic reported that manual review of Claude Code prompts can become habitual, reducing its effectiveness
• ▪Claude Code users approve 97% of permission prompts during manual reviews