Imagine you are using an AI agent for “vibe coding” – writing code. You are tired of manually giving instructions to the AI to do tasks, and want to automate it. However, Claude Code’s “dangerously-skip-permissions” command came with a risk, the AI agent could unknowingly delete your files. Now, Anthropic has a solution, in the form of “auto mode.”
According to the AI startup, the auto mode allows Claude Code to make permission decisions by itself, without having to ask you. This will reduce the need for the AI agent for constant human approval during coding tasks.
The auto mode is pitched as a safer alternative to the “dangerously-skip-permissions” option, which lets the AI operate without any checks but carries significant risks. The new mode aims to balance convenience and control by letting Claude run longer tasks with fewer interruptions while maintaining safety safeguards.
How does Claude Code’s auto mode work?
As the name suggests, Claude Code will be able to do a variety of tasks without needing your manual approval. However, before the AI agent starts a task, it uses an AI-powered classifier to assess the risk level.
The AI then blocks potentially harmful commands such as mass file deletion, sensitive data leaks, or any malicious code execution.
Actions that are deemed safe proceed automatically, while the risky ones are stopped. While Claude Code can then complete the safe tasks on its own, the AI agent may ask for your permission in case the task is deemed risky. This allows you to retain control over sensitive work, while the more mundane tasks can happen on their own.
The auto mode for Claude Code comes at a time when several autonomous coding tools have emerged, from the likes of GitHub and OpenAI, that can execute tasks on a user’s behalf. However, Anthropic’s solution gives AI the power to make decisions for safe tasks.
How to use Claude Code’s auto mode?
Currently, auto mode is available as a research preview for Claude Team users and will soon be rolled out to Enterprise and API customers. It supports Claude Sonnet 4.6 and Opus 4.6 models. Users can enable or disable the mode through settings in the desktop app, command line interface, or VS Code extension.
Anthropic acknowledges that while auto mode reduces risks compared to skipping permissions altogether, it does not eliminate them entirely. The classifier may occasionally block harmless actions or allow risky ones if the context is unclear.
The AI startup has insisted that it plans to refine the system over time, advising users to use the auto mode only in isolated or sandboxed environments to minimise potential damage from unforeseen issues.
Auto mode follows Anthropic’s recent launches of Claude Code Review, an automatic code reviewer, and Dispatch for Cowork. Claude can now also have complete access to your computer, allowing the AI to do tasks for you even when you are away. Anthropic is essentially presenting Claude as not just a tool, but a remote coworker.
