Claude Code Auto Mode: Classifier-Guarded Autonomous Permissions (Research Preview)

Anthropic·high signal

Anthropic launched auto mode for Claude Code on March 24, allowing Claude to handle permission decisions autonomously with a Sonnet 4.6 classifier reviewing each tool call before execution. Blocks destructive actions (mass file deletion, data exfiltration, malicious code) while letting safe operations proceed — a middle ground between manual approval and --dangerously-skip-permissions. Available now for Teams plan users; Enterprise and API access coming soon.