Claude Code Makes Auto Mode Default and Free

Claude Code switched new sessions to auto mode by default on August 14, per Claude Code Docs. A background classifier now approves safe actions on its own. It blocks risky ones instead of stopping to ask for permission on every step. Anthropic also confirmed the classifier's own approval calls stop counting against Pro, Max, and Team plan usage limits. That removes the cost that made people avoid leaving auto mode on in the first place.

A default you already set stays in place. Claude Code offers a one-time prompt to switch you to auto, and you can decline it. Org-managed defaults aren't touched by any of this. Anyone running unattended agents or CI jobs can set permissions.defaultMode: "auto" in ~/.claude/settings.json. That's a safer alternative to --dangerously-skip-permissions, which skips checks entirely instead of enforcing the hard-deny rules auto mode keeps.

The docs don't say how often the classifier pauses anyway, or how it handles actions that fall between clearly safe and clearly risky. That gap matters for anyone scripting heavy file writes or shell commands unattended. It decides how much they can trust auto mode to run without review.

Related Claude Code coverage describes forked subagents becoming the default for child sessions, inheriting the parent conversation and prompt cache automatically.