Claude Code switches to auto mode as its default permission setting on August 14 for Pro, Max, and Team users — the AI can run shell commands without asking each time. Source
What changes:
- Auto mode becomes default for Pro, Max, and Team plans — Enterprise excluded
- Separate classifier reviews shell commands and actions before execution
- 89% dangerous-command catch rate in testing vs. 14% for manual approval
- Users can override — sessions use auto mode unless a different default is set by user or admin
The 14% manual-approval number is the one that matters. Humans clicking through prompts catch almost nothing — fatigue turns the approval gate into decoration. Anthropic’s bet: a classifier doing pattern-matching at machine speed beats a human who stopped reading three commands ago.
NOTE
This ships alongside new weekly usage limits for Claude subscribers. Anthropic estimates 40–80 hours of weekly Sonnet 4 usage depending on project size and auto-accept settings, and says fewer than 5% of users will be affected.
Enterprise plans skip the auto-default — admins in larger orgs want explicit control over what runs on shared infrastructure. Pro and Team users who want the old behavior need to set their permission mode manually.
Available August 14 in Claude Code for Pro, Max, and Team tiers. Override via permission settings to restore manual approval.