Claude Code Switches Auto Mode to Default

Claude Code will apply Auto Mode as the default for new sessions on Pro, Max, and Team plans starting August 14. If users have not set a separate default, a one-time prompt asking whether to switch may appear, and no additional token costs for classifier processing will be charged.

Auto Mode passes all tool calls through a classifier to block irreversible, destructive, or externally targeting actions. When blocked, Claude either finds a safer method or requests user approval; after three consecutive blocks or 20 blocks within a session, it switches to manual approval mode.

Anthropic conducted internal and external red-teaming, prompt injection evaluations, controlled studies with 1,053 paid testers, and analysis of real-world operational sessions. They explained that Auto Mode demonstrated safety equal to or better than manual review across all tested metrics.

97% of manual permission requests were approved, with a rejection rate of only 3%, indicating the possibility that users habitually approve commands without sufficient review. In contrast, the rejection rate for plan approval conversations was 39%.

Auto Mode also supports long-running autonomous tasks. Among enterprises adopting Teams and Enterprise, PR throughput for Auto Mode users increased by approximately 25%, and companies such as Adobe, Nuro, Gusto, and Garner Health are using it as their production default.

For Enterprise, the Claude API, AWS, Amazon Bedrock, Google Cloud Agent Platform, and Microsoft Foundry, it will remain an opt-in feature for the time being. Anthropic plans to collaborate with cloud partners to make it the default in these environments next month, and Enterprise administrators can pre-apply it via managed settings.