Claude Code puts auto mode in the driver's seat

Anthropic, the company behind the AI language model Claude, is set to make auto mode the default setting for its coding assistant from August 14. This change comes after Anthropic's internal testing showed that auto mode is as safe or safer than an average user manually approving each action. Users on Pro, Max, and Team plans will see a one-time prompt to switch if they have a different default already set.

Currently, this feature is an opt-in option for Claude Enterprise, Amazon Bedrock, Google Cloud's Agent Platform, and Microsoft Foundry. Anthropic plans to roll out auto mode to these platforms within the next month. Additionally, Anthropic has eliminated charges for the extra tokens used by the classifier in Pro, Max, and Team plans, and intends to do the same across other platforms.

Auto mode launched in March as a research preview and became generally available on July 10. It serves as an alternative to the default permissions, where every file write and bash command requires manual approval. Auto mode acts as a safety net, sending each tool call through a classifier designed to block irreversible, destructive, or actions outside the user's environment.

If the classifier blocks something, Claude attempts to find a safer alternative. If there are three consecutive blocks or 20 blocks across a session, Claude reverts to manual approvals. During testing, Anthropic found that auto mode blocked 89% of dangerous commands, compared to just 13.6% for an average user manually clicking through prompts.