Anthropic Enables Auto Mode in the Claude Code Programming Tool

Anthropic, one of the leading companies in artificial intelligence, has announced an important update for its Claude Code programming assistant. According to ixbt.com, the company has decided to enable auto mode by default for Pro, Max and Team accounts starting August 14 this year. This change will allow developers to work faster and more efficiently with less human oversight. Techcrunch.com reports .

For context, Anthropic first introduced a trial version of this feature in March. Designed to balance increased speed with control, the mode was developed to make things more convenient for users. From now on, the AI will not stop to request separate human approval for every step.

How auto mode works and how secure it is

According to an official statement published on Friday, Claude Code operating in auto mode continues independently without sending a separate request for every action. The system stops only when an action is deemed "irreversible, destructive or directed outside the environment." This approach significantly speeds up the workflow.

The company also shared interesting statistics from the testing period. According to a study involving 1,053 paid users, auto mode proved significantly safer than manual review. In particular, auto mode detected 89 percent of harmful actions in time, while human-performed manual review achieved a result of just 13.6 percent.

User habits and new protective measures

Experts point out that manual approval can become a habit. According to the statistics, users simply approve as many as 97 percent of permission requests in Claude Code. This reduces the effectiveness of human oversight. Claude Code head Boris Cherny wrote on his page on the social network X (formerly Twitter) that his team had been using only auto mode for many months and could not even imagine returning to permission requests.

To further strengthen security, Anthropic is also introducing a number of additional features. These include screening for prompt injections to prevent data leaks and configurable hard deny rules. These measures ensure that developers can work in a secure and controlled environment.