Anthropic has an important message for Claude users

Anthropic has revealed that its Claude Code platform will switch to auto mode as the default setting from August 14, impacting Pro, Max, and Team accounts. This modification eliminates the need for manual approval of AI actions during programming tasks, allowing Claude Code to process actions automatically unless they are deemed irreversibly destructive or aimed outside the user's environment.

The company first launched auto mode as a test feature in March, demonstrating its superior safety in detecting harmful actions—capturing 89% of potential risks compared to 13.6% identified by human reviewers. Users typically approve most permission prompts manually, making automated oversight more efficient. Anthropic introduced auto mode in March to balance speed and control, finding that it performed 89% better than manual review in catching harmful actions.

The system operates by first checking for existing connector integrations, such as Slack and Google Calendar, and utilizing them if available. If no direct integration exists, Claude Code takes control of the user's screen to perform tasks like opening files, navigating browsers, and typing into forms, always seeking permission before interacting with new applications.

The feature integrates with Anthropic's Dispatch, a task-assignment tool, allowing users to delegate work from their mobile devices, with the results available on their desktops upon arrival. Users can also set recurring tasks, such as scanning emails and compiling weekly metrics. Claude Code Head Boris Cherny expressed his enthusiasm for the feature, stating that he and his team have exclusively used auto mode for months and found it to be a significant improvement over permission prompts.

Anthropic has incorporated additional safety measures, including prompt injection screening and customizable hard deny rules, to prevent data leaks and unauthorized actions. The update promises to streamline workflows and enhance efficiency for developers using Claude Code, reducing the need for constant human oversight while maintaining robust safety controls.