Anthropic has announced a new “auto mode” for Claude, designed to enable more autonomous AI-driven coding while maintaining safety controls. The feature, currently in research preview, allows the model to determine which actions can be executed independently, reducing the need for constant developer oversight.
Auto mode introduces a safeguard layer that evaluates each action before execution, identifying potential risks such as unintended behavior or prompt injection attacks. Safe actions proceed automatically, while higher-risk operations are blocked, aiming to balance efficiency with control as AI systems take on more independent tasks.
The update builds on Anthropic’s broader push into autonomous development tools, including Claude Code Review and Dispatch for Cowork. The feature will roll out to Enterprise and API users and is initially compatible with Claude Sonnet 4.6 and Opus 4.6, with recommended use in isolated environments.
