Anthropic has introduced a new “auto mode” for its Claude AI system, designed to allow the model to independently decide which coding actions can be executed without human approval while applying safeguards to limit risk.
The feature, currently in research preview, reflects a shift toward more autonomous AI tools that can act on behalf of developers without constant supervision.
Balancing Autonomy And Control In AI Coding
Auto mode addresses a common challenge in AI-assisted development, where users must either approve every action manually or allow models to run with minimal oversight.
Anthropic’s approach introduces an intermediate layer, where the AI evaluates each action before execution.
Safe actions proceed automatically, while those flagged as risky are blocked. The system is designed to detect behaviors that were not explicitly requested by the user, as well as attempts at prompt injection, where hidden instructions manipulate the model into unintended actions.
Extension Of Existing Claude Capabilities
The feature builds on Claude Code’s existing command, “dangerously-skip-permissions,” which allows full autonomy but without safeguards.
Auto mode retains the ability for independent execution but adds a review mechanism before actions are carried out.
This shifts decision-making responsibility from the user to the AI, determining when permission is required and when it is not.
Part Of Broader Industry Movement
Anthropic’s update comes amid increasing adoption of autonomous coding tools across the industry.
Companies such as GitHub and OpenAI have introduced systems capable of executing development tasks with limited user input.
Auto mode extends this trend by embedding decision-making directly into the model’s workflow.
Anthropic has not disclosed the specific criteria used by its safety layer to classify actions as safe or risky.
Deployment Scope And Usage Limitations
The feature will be rolled out to enterprise and API users in the coming days.
It is currently compatible with Claude Sonnet 4.6 and Opus 4.6 models. Anthropic recommends that developers use auto mode in isolated or sandboxed environments, rather than production systems, to limit potential impact if unintended actions occur.
The launch follows recent additions to Claude’s developer tools, including automated code review and task delegation features designed to support AI-driven workflows.
Featured image credits: Syllaby.io
For more stories like it, click the +Follow button at the top of this page to follow us.