Auto Mode
Letting an AI coding agent work without approving every step, with a safety classifier watching instead.
Auto Mode is the setting in Claude Code where the agent stops asking permission for each individual action. Rather than a prompt for every command, a classifier inspects what the agent wants to do and interrupts only when something looks dangerous or hard to reverse. Anthropic made it the default for Pro, Max and Team plans in August 2026.
The counterintuitive part is that it tests as safer than asking a human. In Anthropic’s study with 1,053 paid testers, the classifier caught 89 percent of dangerous commands while human reviewers caught 13.6 percent. That is not a comment on developers but on approval fatigue: prompted forty times an hour, people stop reading and start clicking. Anthropic still recommends reviewing changes to production systems by hand, which is the right caveat, since 89 percent is a filter and not a fence.