Claude Code auto mode becomes the default on August 14

Anthropic announced on August 10 that Claude Code's auto mode, which lets the agent execute file writes and bash commands without a manual approval prompt for each one, becomes the default for Pro, Max, and Team plan users starting August 14. Enterprise, API, and cloud-platform users stay opt-in for now, with a default switch planned within a month.

Auto mode still routes every tool call through a classifier trained to block actions that are irreversible, destructive, or aimed outside the working environment. When it blocks something, Claude tries an alternative approach instead of just stopping. Three consecutive blocks, or 20 in a single session, drops the agent back into manual approval mode.

The numbers behind the decision are the real story. In testing across 1,053 participants, human reviewers caught deliberately dangerous commands only 13.6% of the time. The classifier caught 89% of the same set, and blocked all 720 tested attack attempts outright. Anthropic's own telemetry shows why manual review performs that badly: users approve roughly 97% of permission prompts regardless of content, the exact approval-fatigue pattern a separate study published days earlier measured directly.

Anthropic is also dropping the token charge for the classifier itself on Pro, Max, and Team plans, with the same change planned for other tiers. For a team weighing whether to leave auto mode on by default, the classifier's block rate is the number that matters, not whether reviewing every prompt manually feels safer.