Many of my learners set up Claude Code or the Claude desktop app in their first few sessions. So when Anthropic announced on 22 October that auto mode is becoming the default across the Claude apps, I knew the questions would follow. Here is what changes, in plain words.
What auto mode actually does
Until now, when Claude wanted to do something on your behalf (run a command, edit a file, open a website), it stopped and asked you. Allow or deny. Again and again.
In auto mode, it stops asking for every step. Instead, a separate automated check, which Anthropic calls a classifier, reviews each action before it runs:
- Safe actions go ahead. Reading files, routine edits, normal steps in your task.
- Risky actions are blocked. Anthropic says the check targets actions that are irreversible, destructive, or aimed outside your environment.
- Claude looks for another route. When something is blocked, Claude usually finds a safer way or asks you directly.
In Claude Code, if the check blocks Claude 3 times in a row or 20 times in one session, it falls back to asking you, like before.
Who gets it, and when
- Claude Code: auto mode has been the default for new sessions on Pro, Max and Team plans since 14 August 2026.
- Claude app (web, desktop, mobile): Anthropic's 22 October update says auto mode rolls out the following week as the default for new sessions in most Claude apps on Pro, Max and Team.
- Enterprise: stays on Manual by default for now. Anthropic says Enterprise admins will get an email before auto mode becomes the default for their organisation.
If you have already picked a default yourself, Claude Code may show you a one-time prompt asking whether you want to switch. If you pinned a default, nothing changes.
How to switch back to Manual
You are always in control of this setting.
- In the Claude app: look at the permission setting in the message box. Choose Manual and Claude will ask before it takes actions. The setting applies to the whole conversation, and you can change it any time.
- In Claude Code (terminal): press Shift+Tab to cycle through modes. The current mode shows in the status bar.
- In the Claude desktop app for coding: use the mode selector.
- For a whole company: admins can pin a default mode, or switch auto mode off completely, through managed settings.
Why Anthropic made this change
The honest reason is that most of us were not really reading the prompts. Anthropic's data shows users approved 97% of permission prompts in Claude Code. In a test with 1,053 paid testers, people caught only 13.6% of dangerous commands slipped into their prompts, while auto mode caught 89% of the same commands.
I recognise this from my own sessions. After the twentieth "Allow?" box, everyone just clicks yes.
My take: when to use Auto, when to use Manual
In my earlier post on whether it is safe to let AI use your computer, I said keep approvals on and read every prompt. With auto mode, my advice gets a small update, not a reversal.
Auto is fine for:
- Research, summaries and reports
- Drafting documents, slides and spreadsheets
- Organising files in one project folder you gave Claude access to
Switch to Manual for:
- Anything that sends messages or email on your behalf
- Changing or deleting important files
- Payments, bookings, or anything with money
- Work on your company's live systems
Anthropic says the same thing in its own help pages: for work with real consequences, stay close and review what Claude does. Auto mode reduces risk, but it is still a classifier, and Anthropic is clear that it does not eliminate risk.
A simple routine for beginners
- Start your first week on Manual so you see what Claude actually does.
- Move to Auto for research and drafting once you are comfortable.
- Give Claude one project folder, not your whole drive.
- Keep a backup of anything important before a big task.
- Watch the first few minutes of any long task before you walk away.
If you work in a team, write one line in your AI policy: which tasks are allowed on Auto, and which always need Manual. Also see keeping your data private in ChatGPT, Claude and Gemini.