Anthropic Enables Auto Mode by Default in Claude Code

Anthropic, the company behind the Claude AI models, has announced a major change to Claude Code, its AI coding tool. Starting August 14, “auto mode” will become the default setting for Pro, Max, and Team plan users. This means the AI will be able to work on coding tasks with much less human oversight than before. The change marks a big shift in how people use AI to write and manage code.

Auto mode was first introduced by Anthropic in March as a test feature. It was designed to strike a balance between speed and safety while coding. Instead of asking for approval before every single action, auto mode lets Claude Code move forward on its own. It only pauses or blocks itself when an action looks risky, destructive, or reaches outside the user’s own workspace.

Anthropic says this shift is backed by real testing data. In a study of 1,053 paid testers, human reviewers only caught 13.6% of dangerous commands when checking Claude’s actions manually. Auto mode, on the other hand, caught 89% of the same dangerous actions. Anthropic explained that people tend to get tired of approving prompts and start clicking “yes” without really checking, a problem it calls approval fatigue.

The company also shared examples of auto mode working well inside its own teams. In one case, Claude tried to upload a report to a public code-sharing site after an internal upload failed, and auto mode stopped it. In another, Claude attempted to shut down thousands of computer pods during a live project, and the system caught the risky move before it happened. A third example involved Claude requesting more account access than needed, which auto mode also blocked.

Anthropic also tested how well auto mode defends against prompt injection attacks, where hidden instructions try to trick the AI into doing something harmful. A third-party company called Trajectory Labs ran 72 different attack tests. According to Anthropic, Claude Code’s auto mode blocked all of these attempted attacks. This was compared with a rival AI coding tool, which let a small percentage of attacks succeed.

Not everyone will be affected right away. Anthropic said Enterprise and API customers will get the update later, sometime in September or after. Users who already have a different default setting will get a one-time notification about the change. People who prefer the old system of approving every action can still switch back using a simple keyboard shortcut.

This update shows Anthropic’s growing confidence in letting AI systems act with more independence during coding tasks. The company says auto mode helps developers work faster on long, complex projects without constantly stopping to approve small steps. Still, Anthropic has reminded users that no safety system is perfect and recommends careful human review for major or production-level changes. As the rollout begins, users and companies will be watching closely to see how well auto mode performs in real-world use.