Anthropic: Claude Code No Longer Requires Approval for Every Action
Anthropic has announced that Auto Mode will now be the default for Claude Code for Pro, Max, and Team subscribers. A new classification tool allows the AI to autonomously block dangerous commands.
Claude Code. Photo: Geektime
Stop us if you know this scenario: you let Claude Code run, leave the session, and go do something else — only to return and discover that it has barely progressed because it needed your approval for some action. Now, Anthropic is bringing an interesting change to its popular tool to address this issue.
Better than you, and you know it
Over the weekend, Anthropic announced a significant change to Claude Code: it will stop asking for your approval for every single action. The company is enabling Auto Mode by default for Pro, Max, and Team subscribers starting this week. Users who previously set a different default may receive a one-time prompt to confirm if they would like to switch to the automatic mode permanently.
Until now, Claude Code ran by default in manual mode, requiring you to approve every change to ensure security and catch bugs in real-time. In practice, for those who do not write code daily, this is a mode where users likely just clicked 'Accept All' automatically.
Why would Anthropic switch to an automatic mode that might be considered riskier? According to the company, the change comes because it has managed to make the tool much smarter and more cautious. Instead of prompts, the automatic mode runs every action through a special classification tool that blocks the execution of irreversible, destructive actions or those aimed at working outside the environment where the code runs.
The numbers speak for themselves
Anthropic notes that in recent months they have worked hard on testing the capabilities of the automatic mode. After Red Teaming processes, Prompt Injection tests, and research involving more than 1,000 human participants, they feel confident enough to make it the default.
Research results showed that while the automatic mode blocked 89% of dangerous commands, human testers blocked these malicious commands in only 13.6% of cases.
The default change will begin on August 14, and every user will be able to change the setting as they see fit. However, Anthropic clarifies that as of now, access via API, Enterprise plans, and access to Claude Code through cloud services such as AWS, Azure, and GCP will not change the tool's default.
