Definition
Plain language
Claude Code's setting that lets the coding assistant run unattended, with a second model checking each step instead of a human.
As stated in the literature
Anthropic's Claude Code feature that replaces per-action human confirmation with a staged automated monitor over the serialized session transcript; the subject of the red-teaming study in this episode.
Also called: Auto-Mode
Why it matters: It is the difference between an assistant that needs babysitting and one that can finish long jobs alone, which also means the automated reviewer is now the only thing standing between a bad instruction and your machine.
For example, a developer can hand Claude Code a long refactoring job before going to lunch, and instead of stopping to ask permission for each file edit or shell command, a separate model reviews each step and decides whether to let it through.
Heard on the show
“Production Auto Mode swapped its monitor model partway through the study, and the mean attack success barely moved.”Episode 269 — How a Forged Transcript Got Model Weights Past a Safety Monitor