Anthropic makes Claude Code's auto mode the default

Anthropic will make auto mode the default setting for Claude Code, its AI coding tool, across Pro, Max, and Team subscription tiers starting August 14. The company first introduced a test version of the feature in March, pitching it as a way to balance speed with human oversight. Today, when Claude Code operates in its default mode, it pauses at each step to ask for a person’s approval before taking an action. In auto mode, that pausing goes away: the tool proceeds on its own unless it judges a given action to be irreversible, destructive, or aimed outside the user’s own environment, in which case it still stops to ask.

To justify the shift, Anthropic pointed to results from a study of 1,053 paid Claude Code testers comparing auto mode against manual, step-by-step review. In that test, auto mode caught 89% of harmful actions before they were carried out, while human reviewers manually approving each step caught only 13.6% of the same actions. Anthropic attributed part of that gap to habit: it said users approve 97% of the permission prompts Claude Code shows them, regardless of what the prompt actually asks.

Boris Cherny, who leads Claude Code at Anthropic, said in a post on X that he and his team have relied on auto mode exclusively for months already and would not want to go back to manual permission prompts. Alongside the default change, Anthropic said it has also added new safety features meant to keep unsupervised auto mode in check, including screening for prompt-injection attempts and customizable hard deny rules designed to block actions such as data exfiltration.

Why it matters

The change pushes Claude Code further toward autonomous operation, making Anthropic’s own safety testing — rather than a person at each step — the primary check on what the tool is allowed to do in a user’s codebase.