AI autonomous decision-making power expands, Anthropic introduces automatic mode for Claude Code

```

Anthropic is giving its AI programming tool greater autonomy while trying to balance efficiency and safety.

On March 24, Anthropic announced the launch of "auto mode" for Claude Code, allowing the AI to determine which operations can be executed directly without waiting for individual user approval.

This feature is currently available to team plan users as a research preview and will be rolled out to enterprise and API users in the coming days.

The core of this new feature lies in its built-in safety mechanism: before any operation is executed, it must pass an AI safety review. The system will automatically allow operations deemed safe and block those that pose risks.

Anthropic stated that this safety layer can also detect prompt injection attacks, where malicious commands are hidden in content processed by the AI in an attempt to induce the model to perform unintended actions.

The company advises users to use this new feature in isolated sandbox environments to prevent potential risks from spreading to production systems.

Developer pain points drive product iterations

For developers currently using AI programming tools, a common dilemma is either supervising every step of the AI's operations, or letting the model run autonomously and accepting unforeseeable risks.

Anthropic's auto mode is essentially an upgrade and extension of Claude Code's existing "dangerously-skip-permissions" command, which no longer asks the user for confirmation.

This command originally delegated all decision-making to the AI, while the new mode adds a layer of safety filtering on top of this.

By letting the AI rather than the user decide when to request permissions, Anthropic is attempting to provide developers with higher levels of security without sacrificing execution efficiency.

Companies like GitHub and OpenAI have already launched autonomous programming tools that can carry out tasks for developers. Anthropic’s latest move takes this further by shifting control over permission decisions from the user to the AI itself.

The release of auto mode comes on the heels of a series of recent Anthropic product updates, including Claude Code Review—which automatically detects bugs before code merges—and Dispatch for Cowork, which allows users to assign tasks to AI agents.

This series of moves shows that Anthropic is systematically building an autonomous AI workflow product matrix for enterprise developers.

Key details yet to be clarified

Nevertheless, there remain notable uncertainties regarding this feature.

Anthropic has not yet publicly disclosed the specific standards by which its safety layer determines risk levels for operations. This is precisely the key information developers need to understand before adopting the feature at scale.

Additionally, auto mode currently only supports Claude Sonnet 4.6 and Opus 4.6 models, and is still in the research preview stage, meaning the product has not been finalized.

For enterprise users considering deploying this feature in production-adjacent environments, these limitations and lack of transparency may be important factors in their evaluation.

Risk Disclosure and DisclaimerThe market entails risks, and investment should be undertaken with caution. This article does not constitute personal investment advice and does not consider any user’s particular investment objectives, financial situation, or needs. Users should consider whether any opinions, views, or conclusions in this article fit their specific situation. Investments made based on this article are at your own risk. ```