Author: Anthropic
Compiled by: Deep Tide TechFlow
Deep Tide Guide: Claude Code previously required manual approval for each step or used --dangerously-skip-permissions to completely bypass checks.
Auto Mode is a middle path: the AI autonomously determines whether each action is safe, automatically intercepting dangerous actions, allowing longer tasks to be completed without intervention.
Currently open to Team plan users as a research preview.
The full text is as follows:
Auto Mode of Claude Code
Today, we launched Auto Mode— a new permission mode of Claude Code. In this mode, Claude makes permission decisions on your behalf and checks each action through safety mechanisms before execution. Currently open to Team plan users as a research preview, it will be rolled out to Enterprise plan and API users in the coming days.
How it Works
The default permission settings of Claude Code are intentionally conservative: each file write and bash command execution requires your manual approval. This is a safe default, but at the cost of being unable to start a large task and walk away, as Claude frequently asks for manual confirmation during the process. Some developers choose to skip all permission checks using --dangerously-skip-permissions, but this may have dangerous and destructive consequences and should not be used outside isolated environments.
Auto Mode is a middle path: it allows you to run longer tasks with less interruption while introducing less risk than skipping all permissions. Before executing each tool call, a classifier reviews it to detect potential destructive operations, such as bulk deletions of files, sensitive data leaks, or executing malicious code.
Operations deemed safe by the classifier will be executed automatically, while risky operations are intercepted, guiding Claude to use alternatives. If Claude insists on executing operations that are repeatedly intercepted, this will ultimately trigger a permission prompt initiated towards the user.
Usage Expectations
Compared to --dangerously-skip-permissions, Auto Mode reduces risk but doesn't completely eliminate it, and we still recommend using it in isolated environments. The classifier may still allow certain risky operations, such as when the user's intent is ambiguous, or when Claude lacks sufficient context about your environment to assess whether an operation may carry additional risk. It may also occasionally intercept harmless operations. We will continue to improve this experience.
Auto Mode may have a slight impact on token consumption, costs, and latency for tool calls.
How to Get Started
Auto Mode is currently open to Claude Team users as a research preview and will be rolled out to Enterprise and API users in the coming days. It supports Claude Sonnet 4.6 and Opus 4.6.
Administrators: Auto Mode will soon be available for all Claude Code users on Enterprise, Team, and Claude API plans. To disable it in the CLI and VS Code extension, set `"disableAutoMode": "disable"` in the hosting settings. The Claude desktop application has Auto Mode turned off by default, which can be enabled through organization settings → Claude Code.
Developers: Run `claude --enable-auto-mode` to enable Auto Mode and then press Shift+Tab to switch to this mode. In the desktop application and VS Code extension, first enable Auto Mode in settings → Claude Code, then select it in the permission mode dropdown in the session.
免责声明：本文章仅代表作者个人观点，不代表本平台的立场和观点。本文章仅供信息分享，不构成对任何人的任何投资建议。用户与作者之间的任何争议，与本平台无关。如网页中刊载的文章或图片涉及侵权，请提供相关的权利证明和身份证明发送邮件到support@aicoin.com，本平台相关工作人员将会进行核查。
