Anthropic turns Claude Code's auto mode on by default

Anthropic is making auto mode the default for Claude Code on Pro, Max, and Team accounts starting August 14, reducing the number of permission prompts developers see during AI-assisted coding sessions. The company first tested the feature in March as an optional balance between speed and control.

In auto mode, Claude Code proceeds with actions unless they are deemed "irreversible, destructive, or aimed outside your environment." Anthropic argues this is both faster and safer than manual review. In a study with 1,053 paid testers, auto mode caught 89% of harmful actions, compared to just 13.6% for human review. The company attributes the gap partly to habituation: users approve 97% of permission prompts in Claude Code, meaning the checks become routine rather than meaningful.

Claude Code Head Boris Cherny has been using auto mode exclusively for months and says he can't imagine going back to permission prompts. Anthropic has also added safety features such as prompt injection screening and customizable hard deny rules to prevent data exfiltration, reducing the risk of automated actions going wrong.

What this means for developers

For developers who rely on Claude Code, this shift removes a frequent point of friction. Instead of babysitting every action, you can let Claude run with guardrails that catch genuinely risky operations. The tradeoff is trust: you're delegating more authority to the model, and the safety stats, while impressive, come from Anthropic's own testing.

The bigger implication is the direction of AI coding tools: as models get more reliable, the default is moving from "ask first" to "act first, flag anomalies." Teams will need to configure hard deny rules carefully and audit behavior, especially when working with sensitive codebases or external services.