Anthropic will turn on Claude Code auto mode by default on Aug. 14

In short: Anthropic says it will make Claude Code’s “auto mode” the default setting for Pro, Max, and Team accounts starting August 14.

What happened

Anthropic makes an AI assistant called Claude, and it also offers Claude Code, a tool that helps people write and edit software. The company says Claude Code will soon do more steps on its own without asking for permission each time.

Auto mode was first tested in March. In this mode, Claude Code keeps going unless it thinks an action is “irreversible, destructive, or aimed outside your environment.” In plain terms, it is like letting a helper keep working without tapping you on the shoulder for every small decision, but still stopping before something that cannot easily be undone.

Anthropic says its testing found auto mode was safer than relying on people to approve each step manually. In a study with 1,053 paid testers, the company says auto mode flagged 89% of harmful actions, while human review flagged 13.6%. Anthropic also said users approved 97% of permission prompts, suggesting people often click “allow” out of habit.

The company says it has been adding safety features too. It mentioned “prompt injection” screening, which is a way to catch tricky instructions that try to hijack the tool (like a stranger slipping new directions into a to do list). It also mentioned customizable “hard deny” rules to help prevent data exfiltration, meaning data being copied out somewhere it should not go.

Why it matters

Tools like Claude Code can make changes to real projects, so defaults matter. Turning on auto mode by default could speed up work, but it also increases the importance of clear guardrails, since more actions will happen with less human checking.

Source: TechCrunch AI