Anthropic is testing a new “auto mode” for Claude Code that lets the AI decide which actions are safe to take on its own, without waiting for human approval — but with a safety layer to block risky behavior.
The update aims to streamline AI coding by reducing the need for constant oversight, while still preventing unintended actions, including prompt injection attacks. Safe actions proceed automatically, while risky ones are blocked.
Currently in research preview, auto mode is available for Enterprise and API users with Claude Sonnet 4.6 and Opus 4.6, and Anthropic recommends using it in sandboxed environments to limit potential issues. This comes alongside the company’s other AI tools like Claude Code Review and Dispatch for Cowork, pushing further into autonomous coding territory.
