Claude Code turns on auto mode by default

Starting August 14, Claude Code turns auto mode on by default for Pro, Max, and Team accounts. Instead of a permission prompt before each action, every tool call routes through a classifier that blocks anything “irreversible, destructive, or aimed outside your environment”; when it blocks something, Claude typically finds a safer way to proceed rather than stopping. Enterprise, API, and platform users (Bedrock, Vertex, Foundry) stay opt-in for now, with Anthropic planning to extend the default there within a month.

Anthropic’s case rests on its own numbers: in a controlled study of 1,053 paid testers, auto mode caught 89% of dangerous commands against 13.6% for manual human review, a gap it attributes to approval fatigue: testers approved 97% of prompts reflexively. Apollo Research red-teaming cut the classifier’s miss rate from 12% to 7% after hardening, and new hard-deny rules specifically block data exfiltration and check git push destinations before executing. Adobe, Nuro, Gusto, and Garner Health are cited reporting roughly 25% more merged PRs among auto mode adopters.

What it means for you

This goes further than auto mode’s July rollout on Bedrock and Vertex, which only touched enterprise cloud platforms; this one hits every individual Pro, Max, and Team user by default. If you lean on approval prompts as an actual safety control rather than a formality, that control disappears August 14 unless you act first: press Shift+Tab in the CLI, use the mode dropdown on desktop, or have an admin set defaultMode or disableAutoMode. For the broader tradeoff between agent autonomy and operator control, see running coding agents remotely and when you let an AI spend money.