Auto mode is now the default in Claude Code for Pro, Max, and Team plans

Anthropic is making auto mode the default in Claude Code for Pro, Max, and Team plans starting August fourteenth, confident that automatic action approval is safer than manual review. According to Simon Willison's Weblog, internal testing showed compelling evidence: in a study with over one thousand paid testers, only thirteen point six percent refused to approve a harmful command when prompted, while auto mode would have blocked eighty-nine percent of such actions. Anthropic also commissioned Trajectory Labs to test seventy-two prompt injection scenarios; the evaluation found zero successful attacks out of seven hundred twenty attempts against the latest Claude models running auto mode. However, Willison notes significant gaps remain—the evaluation couldn't cover all attack vectors, and he worries about sophisticated threats like malicious packages designed to exfiltrate data while appearing legitimate.

Listen to this story

Hear this and more stories in a personalized audio briefing.

Open The Chonkerton