Starting August 14, Pro, Max, and Team accounts will get auto mode by default, with new safety guardrails.

Anthropic will switch Claude Code to auto mode by default for Pro, Max, and Team accounts on August 14. Auto mode lets the coding agent execute steps without asking for approval at every prompt. The agent pauses only for actions the system deems irreversible, destructive, or aimed outside the user’s environment.

Anthropic first tested auto mode in March and now rolls it out after a study of 1,053 paid testers. The company reports auto mode caught 89% of harmful actions while human review caught only 13.6%. Claude Code Head Boris Cherny said his team has used auto mode exclusively for months and would not return to permission prompts.

For builders and operators, the shift means fewer interruptions during long coding sessions, but it also requires trust in automated safeguards. Manual review has a catch: users approve 97% of permission prompts, which can make review habitual and less effective. New safety features include prompt injection screening and customizable hard deny rules to block data exfiltration.

Starting August 14, Pro, Max, and Team users will see auto mode as the default in Claude Code. Teams should review their hard deny rules and test prompt injection screening before the switch. Anthropic’s study suggests automation can outperform human review, but real-world deployments will determine if the safety checks hold up.

What matters

• Anthropic defaults Claude Code to auto mode for Pro, Max, and Team accounts on August 14.
• Builders get faster code workflows but must trust auto mode’s risk checks for irreversible actions.
• Watch for prompt injection screening and custom hard deny rules to block data exfiltration.

Why it matters

Watch for prompt injection screening and custom hard deny rules to block data exfiltration.

This GenAI News article was prepared in original wording using reporting and materials published by TechCrunch AI. Source reference: https://techcrunch.com/2026/08/09/anthropic-is-turning-claude-codes-auto-mode-on-by-default/.

Drafted by the GenAI News review pipeline.