Anthropic's Claude Code auto mode becomes default, catches 89% of dangerous commands

Anthropic is rolling out a major change to Claude Code, its AI-powered coding assistant, by making auto mode the default for all Pro, Max, and Team paid plans starting August 14. Instead of asking users for approval on every tool call, auto mode routes each command through a classifier that automatically blocks irreversible or destructive actions—such as deleting production databases or executing arbitrary shell scripts without review. The shift is designed to reduce friction while maintaining safety, allowing developers to move faster without constantly interrupting their workflow with permission prompts.

The decision is backed by a controlled study with 1,053 paid testers, which found that auto mode caught 89% of dangerous commands compared to just 13.6% caught by humans relying on traditional approval prompts. That 89% detection rate comes from the classifier's ability to recognize risky patterns in real time, while humans often miss subtle destructive operations buried in long command sequences. Teams using auto mode also shipped roughly 25% more pull requests, indicating a significant productivity boost. Enterprise customers and API users will retain the ability to opt in manually, giving organizations with strict compliance needs time to evaluate the new safety classifier before adopting it by default. Anthropic says auto mode is part of its broader effort to make AI agents safer and more autonomous, and the company expects to expand the feature to enterprise plans after further testing.