Developers who pay for Claude Code will soon spend less time approving actions. Starting August 14, Anthropic is making auto mode the default for Pro, Max, and Team accounts, after testing the feature since March.
In auto mode the coding assistant proceeds without asking permission at every step, pausing only for actions it deems irreversible, destructive, or aimed outside the user’s environment. Anthropic says the approach is actually safer than manual oversight, citing a study of 1,053 paid testers in which auto mode stopped 89% of harmful actions while human review caught only 13.6%. Manual review, the study found, has become habitual, with 97% of permission prompts approved anyway.
Boris Cherny, who leads Claude Code, said on X that his team has relied on auto mode exclusively for months and could not imagine returning to permission prompts.
Anthropic is coupling the change with fresh safety measures, including prompt injection screening and customizable hard deny rules aimed at blocking data exfiltration. The tool is moving toward autonomous execution, with built-in guardrails replacing the approval clicks developers once made at every step.
