Claude Code programming will soon necessitate even less human intervention, as Anthropic announces it will switch auto mode to the default setting for Pro, Max, and Team accounts beginning August 14.

The company initially introduced a test version of auto mode in March, promoting it as a mechanism to strike a balance between speed and oversight. As Anthropic detailed in its announcement on Friday, when Claude Code operates in auto mode, it will continue without asking for human consent at each step unless an action is identified as “irreversible, destructive, or aimed outside your environment.”

Anthropic further stated that during testing, auto mode was found to be safer than manual review — in a study involving 1,053 paid testers, auto mode identified 89% of harmful actions, while human review detected only 13.6%. (This might be due to the fact that “manual review can become routine: users approve 97% of permission prompts in Claude Code.”)

In a post on X, Claude Code Head Boris Cherny remarked, “The team and I have been using Auto mode exclusively for many months. I can’t envision returning to permission prompts!”

The company also mentioned it is incorporating new safety features such as prompt injection screening and customizable hard deny rules to mitigate issues like data exfiltration.