Anthropic makes auto mode Claude Code's default

Anthropic will make auto mode the default permission setting in Claude Code for Pro, Max and Team plans starting 14 August, ending the requirement to approve most actions individually. Under auto mode, Claude Code proceeds without asking unless an action is irreversible, destructive or reaches outside the user's environment. Users who have already selected a different default, or pinned one, keep their existing setting unless they choose to switch.

The change follows testing with more than a thousand paid users, in which the auto mode classifier caught 89 percent of harmful actions, comfortably ahead of the rate caught under manual review in the same trials. Anthropic said it will stop charging Pro, Max and Team users for the extra tokens the classifier consumes per tool call. The feature stays opt-in for Enterprise, the API and cloud platforms including AWS, Google Cloud and Microsoft Foundry.

Shifting the default from asking permission to reviewing after the fact changes what a coding agent has to get right the first time, and puts more weight on the classifier catching mistakes than on the user catching them.