Anthropic Defaults Claude Code Auto Mode for Paid Users, Citing 89% Dangerous-Command Catch Rate
Updated
Updated · InfoWorld · Aug 11
Anthropic Defaults Claude Code Auto Mode for Paid Users, Citing 89% Dangerous-Command Catch Rate
3 articles · Updated · InfoWorld · Aug 11
Summary
August 14 marks the switch to auto mode by default for new Claude Code sessions on Pro, Max and Team plans, with Enterprise, API and cloud users due within a month.
97% of permission prompts were already being approved and 62% of active CLI users had used Bash bypasses or “don’t ask again,” pushing Anthropic to replace constant approvals with a safety classifier.
1,053 paid professional testers found the default auto mode caught 89% of deliberate dangerous commands versus 13.6% for human reviewers; after three consecutive blocks or 20 in a session, Claude falls back to manual approvals.
Analysts said the change should let developers hand off longer tasks with fewer pop-ups, but it may add latency to routine work and make enterprise oversight harder when agents run unattended for hours.
Anthropic is keeping user controls in place—pinned preferences stay unchanged and some users will get a one-time prompt—while covering the extra token cost from classifier checks for Pro, Max and Team users.