Anthropic makes auto mode the Claude Code default on Pro, Max, and Team plans August 14

From August 14, new Claude Code sessions on Pro, Max, and Team plans route each tool call through a classifier that blocks irreversible or destructive actions instead of prompting; in a 1,053-tester study it caught 89% of dangerous commands against 13.6% for human review. Classifier token overhead is no longer billed on those plans, and broad allow-rules such as Bash(python:*) are set aside while auto mode runs — Enterprise, API, and cloud-partner deployments stay opt-in for now.

Source: claude.com ↗

users approve 97% of permission prompts in Claude Code

Anthropic

Why this matters

→ Auto mode catches 89% of dangerous commands; humans approve 97% of prompts reflexively
→ Longer uninterrupted work sessions ship ~25% more PRs across Teams/Enterprise
→ No more classifier token costs on Pro, Max, and Team plans

Automation over interruption