Anthropic is preparing to make Claude Code more independent by introducing automatic approval as the default setting for users on its Pro, Max and Team plans.

Beginning August 14, the company plans to activate Auto mode by default, reducing the number of times developers must manually approve actions while using Claude Code.

Anthropic initially introduced Auto mode as an experimental feature in March. The system is designed to allow Claude Code to continue working without repeatedly stopping to request permission, except when it identifies an operation as irreversible, destructive or outside the user’s working environment.

The company says the approach could improve both productivity and security. During testing involving 1,053 paying users, Anthropic reported that Auto mode detected 89 percent of harmful actions, compared with 13.6 percent detected through manual review.

Anthropic suggested that one reason for the difference may be that users frequently approve permission requests without closely examining them. According to the company, users approve approximately 97 percent of permission prompts presented by Claude Code.

Boris Cherny, head of Claude Code, has also expressed strong confidence in the feature. He said that he and his team have relied exclusively on Auto mode for several months and have found it difficult to imagine returning to the previous permission-based workflow.

At the same time, Anthropic says it is strengthening the system’s security architecture. New measures include screening for prompt injection attacks and customizable rules that can automatically block particularly sensitive operations, including attempts to remove or transfer data from a user’s environment.

The development reflects the broader shift in AI-assisted programming toward systems capable of carrying out increasingly complex tasks with less direct human intervention.

Source: TechCrunch