Anthropic Makes Auto Mode Default for Claude Code to Boost AI Agent Safety

Anthropic announced that starting August 14, it will make Auto Mode the default permission setting for Claude Code, affecting Pro, Max, and Team subscribers. Auto Mode uses a separate AI classifier to check tool calls and shell commands in real time, allowing safe actions while blocking harmful ones. This change aims to balance AI agent autonomy with safety.

In tests, Auto Mode caught 89% of dangerous commands, much better than the 13.6% caught by old manual approval. In prompt injection tests, Claude Fable 5, Opus 5, and Sonnet 5 stopped all attacks, while GPT-5.6 Sol had a 5.83% success rate in Codex's auto-review mode.

Anthropic plans to extend Auto Mode to Claude Enterprise, API, and AWS platforms within a month, and remove related costs. This move could raise safety standards for AI agents and affect how developers use them.