Claude’s apps start in auto mode next week. What it does, and how to switch back

What changes next week

On October 22, 2026, Claude’s account on X announced: “Starting next week, sessions across Claude web, desktop, and mobile apps will run in auto mode by default on Pro, Max, and Team plans. Rather than asking you to approve each step, an automated check blocks risky actions. Switch back to manual anytime.”

The same day, Anthropic added an update to its article on auto mode: auto mode “is rolling out next week as the default for new sessions in most Claude apps on Pro, Max, and Team plans.” Neither names a day, and neither says which apps “most” leaves out. Next week begins Monday, October 26.

The setting is the one in the message box where you hand Claude a task. Anthropic is merging chat and Cowork into one conversation, in stages, and its help center already lists that setting as “Auto (default on Pro, Max, and Team plans)” and “Manual (default on Enterprise plans).” According to the Cowork guide, auto mode “applies to all of your existing connectors, plugins, the built-in browser, Claude in Chrome, and some Cowork actions like fetching websites.”

What auto mode does

The Claude apps have three permission modes. Anthropic’s Cowork guide describes them:

• Manually approve (Manual), formerly “Ask before acting.” Claude pauses before actions, and you allow or deny each one.
• Automatically approve (Auto). Claude “reviews each action for safety (such as checking for data exfiltration or prompt injection) and automatically blocks anything it determines to be unsafe.” After a block, it looks for a safer way to finish the task or asks you, and if it keeps running into blocks, it goes back to asking at every step.
• Skip all approvals (Skip), formerly “Act without asking.” Nothing checks Claude’s actions. Anthropic says to use it only “when you completely trust every action, connector, file, app, etc. involved in the task.”

Each mode treats the permission you set on a connector’s tools differently:

Connector tool set to	Manual	Auto	Skip
Always allow	Approved	Read-only tools approved; Claude decides on write and delete tools	Approved
Needs approval	Asks you	Claude decides	Approved
Blocked	Denied	Denied	Denied

The middle row is the one to notice. In Auto, a tool you set to “Needs approval” no longer stops to ask you: the check decides. Only “Blocked” holds in every mode. Deleting files is handled on its own: by default, Claude asks before permanently deleting any files.

Where auto mode is already the default

Claude Code has started in auto mode since the summer. It became the default there for Pro, Max and Team on August 14, as Anthropic’s ClaudeDevs account announced, and Claude Code’s documentation now says that from version 2.1.283 it’s the starting mode for terminal and VS Code sessions “on every plan and provider.”

Where	Pro, Max and Team	Enterprise
Claude apps on the web, desktop and mobile	Auto, from next week	Manual; admins get an email before it changes
Claude Code in a terminal or VS Code	Auto since August 14	Auto from version 2.1.283
Claude Code’s claude -p and the Agent SDK	Manual	Manual

Admins keep the last word. In the apps, the Cowork guide says that on Team and Enterprise plans, “your admin controls whether ‘Automatically approve’ is available to your organization and whether new tasks start in it.” If it’s off, Auto doesn’t appear in your mode selector. In Claude Code, admins do the same with disableAutoMode in managed settings.

How safe it is

Anthropic’s case rests on how people really answer prompts. In Claude Code, users approve 97% of permission prompts, and Anthropic says “an approval rate that high suggests many users are clicking through reflexively rather than reviewing each command.”

In a study with 1,053 paid testers, Anthropic swapped one permission prompt in each session for a clearly dangerous command. People caught it 13.6% of the time (143 of 1,053), and auto mode blocked 89% (937 of 1,053). People got worse as a session went on, from about 17% early to about 5% after 50 or more prompts.

The limits are part of the answer:

• The figures are for Claude Code. Anthropic’s update says: “The data in this article below is specifically for Claude Code.” It adds that its testing was used “to strengthen how auto mode works across all our products.”
• It misses things. When auto mode launched in March, Anthropic’s engineering write-up measured a 17% miss rate on real cases of Claude going past what the user had authorized, and called it “the honest number.” Against synthetic attacks from the safety firm Apollo Research, the miss rate has since fallen from 12% to 7%, a figure Anthropic says “should not be taken as the miss rate on real traffic.”
• Anthropic says so itself. Auto mode “relies on classification systems and therefore does not eliminate risk.” Its Cowork guide adds: “For work with real consequences—money, messages sent as you, important files—stay close and review what Claude does or consider switching back to ‘Manually approve.’”

How to switch back to Manual

1. In the Claude apps, open the permission setting in the message box and choose Manual. You can change it at any time, and it applies to the whole conversation.
2. In Claude Code’s terminal, press Shift+Tab: from auto, the first press switches to Manual. To start every session in Manual, set defaultMode to "default" under permissions in ~/.claude/settings.json. Claude Code may ask once whether to switch it to auto; decline, and your setting stays.
3. In the desktop app’s Code tab, use the mode selector under the prompt box. It remembers your choice for each folder.
4. For a whole organization, a Team or Enterprise admin can turn Auto off, in the apps and in Claude Code.

What Anthropic hasn’t said

• The day. “Next week” is all it has said.
• Which apps are left out. The update says “most Claude apps” without naming the rest.
• Whether Manual sticks. The update covers “new sessions.” Anthropic hasn’t said whether the apps will remember a choice of Manual for the next session, the way Claude Code keeps a default you’ve set.
• Whether the check uses your limits. Since August 7, Anthropic hasn’t charged Claude Code users on Pro, Max and Team for the classifier’s extra tokens. For the apps, its help center says only that “everything you do with Claude counts toward your plan’s usage limits.”
• When Enterprise changes. Only that admins get an email first.

Choosing a mode for each project

Some work deserves Auto and some deserves a person reading every command, often in the same week. Enjoy runs Claude Code on the Claude plan you already have, with Claude Code’s own modes, Auto among them and the default, so the checks are Anthropic’s. Enjoy remembers the mode you last used with each agent in each project, so a production repository can stay on Manual while a side project runs in Auto. When Claude asks, the request appears as a card with the exact command, to allow once or decline. Enjoy never picks a broader mode for you, and Codex and the other agents keep their own modes beside it. Enjoy itself is free.

We’ll update this memo when auto mode reaches the apps, or when Anthropic sets a date for Enterprise.