The short answer
To turn off Cowork in Anthropic's Claude for a Team or Enterprise organisation, an Owner opens Organization settings, then Cowork, and switches off Enable for your organization. For Claude Desktop on your own inference provider, set the desktopHome key to off: from 12:00 PM PT on 10 November 2026, an unset desktopHome reads as standard and turns Cowork on, even where coworkTabEnabled is false. The value off turns chat off too.
| Made by | Anthropic |
|---|---|
| Change date | 10 November 2026, 12:00 PM PT (20:00 UTC) |
| Who it affects | Claude Desktop on your own inference provider (3P) with desktopHome unset |
| Team and Enterprise switch | Organization settings, Cowork, Enable for your organization (Owner or Primary Owner) |
| desktopHome values | standard (chat and Cowork on) or off (chat and Cowork off, Claude Code unaffected) |
| Deprecated keys | coworkTabEnabled, chatTabEnabled, chatAdvancedFileAnalysisEnabled: warning from 17 November, ignored from 17 December 2026 |
| Personal plans | No off switch found; the new experience cannot be switched back |
What changes on 10 November 2026
Claude Cowork is Anthropic's mode for handing Claude a whole task: it can read folders, run code in an isolated virtual machine on the user's computer or in the cloud, use a browser and hand back finished files. Since 16 September 2026 Anthropic has been merging chat and Cowork into one Claude, which its Claude Desktop documentation calls Unified Claude. Some IT and security teams are not ready for every employee to have an agent that runs code and uses the web, so the question of how to turn Cowork off comes up quickly.
The deadline that matters is for organisations that run Claude Desktop on their own inference provider, which Anthropic's docs call third-party or 3P deployments: Claude Desktop pointed at Amazon Bedrock, Google Cloud's Agent Platform, Microsoft Foundry, a compatible gateway or the Anthropic API instead of users signing in to claude.ai. Anthropic's Claude Desktop changelog marks this as a breaking change. When a current version of Claude Desktop starts after 12:00 PM Pacific Time (20:00 UTC) on 10 November 2026, it reads an unset desktopHome setting as standard, which turns on Unified Claude. Anthropic's docs say Cowork is then on even where you had turned it off with the older coworkTabEnabled setting set to false.
In other words, if your configuration relies on the old switches, Cowork comes back on 10 November unless you act.
Which situation are you in?
| Your setup | Where Cowork is controlled | Does 10 November affect you? |
|---|---|---|
| Claude Team or Enterprise on claude.ai | Organization settings, then Cowork, toggle Enable for your organization | Anthropic's 10 November notice is written for Claude Desktop 3P configuration; the claude.ai toggle is documented separately |
| Claude Desktop on your own provider (3P), set up through MDM or a configuration file | The desktopHome key | Yes, if desktopHome is unset |
| Claude Desktop 3P managed in the Enterprise Admin Console | Organization settings, Desktop 3P, Capabilities, Allowed surfaces | Yes, if the unified view switches are not set |
| Personal Pro, Max or Ultra account | No off switch found in Anthropic's help centre | The new experience reaches these plans first, and you cannot switch back |
How to turn off Cowork on a Claude Team or Enterprise plan
Anthropic's help article on Cowork for Team and Enterprise plans gives these steps. Cowork is on by default, except in organisations that have enabled HIPAA, where it starts off.
- Sign in to your Team or Enterprise organisation as an Owner or Primary Owner.
- Open Organization settings, then Cowork.
- Find the Enable for your organization toggle under Cowork.
- Turn it off. Cowork is then unavailable to everyone in the organisation.
This is an organisation-wide switch. On Enterprise plans you can use groups and custom roles to give Cowork to specific teams only. If your worry is where the work runs rather than whether it runs, there is a separate toggle on the same page, Run Cowork in the cloud: on by default for Team, off by default for Enterprise. Turning it off keeps Cowork on users' own machines. On Team (in beta) and Enterprise, the unified chat and Cowork experience appears only where Chat and Cowork unified is turned on.
How to keep Cowork off in Claude Desktop 3P deployments
For 3P deployments, the key that matters from now on is desktopHome, which Anthropic's release notes added in version 2.31226.0 on 8 October 2026. It takes exactly two values:
- standard turns on Unified Claude, with chat and Cowork in one view.
- off turns Chat and Cowork off entirely. Claude Code is not affected; the separate isClaudeCodeForDesktopEnabled setting still decides that.
There is no value for chat without Cowork. If you want chat with a more limited agent, Anthropic points to other settings instead: disabledBuiltinTools (adding Bash to it stops Claude running shell commands), builtinToolPolicy, allowedWorkspaceFolders and scheduledTasksEnabled. Those settings narrow what Cowork can do; they do not hide it. Note also that the value is case-sensitive: Anthropic says a value the app does not recognise, such as Standard with a capital S, is read as off.
With MDM or a configuration file
- Decide on standard or off. Leaving desktopHome out is the one choice that changes on 10 November.
- On macOS, set desktopHome in your MDM profile for com.anthropic.claudefordesktop. A .mobileconfig profile delivered by MDM lands in /Library/Managed Preferences automatically.
- On Windows, write it as a REG_SZ value directly under HKLM\SOFTWARE\Policies\Claude, which Anthropic recommends, or HKCU\SOFTWARE\Policies\Claude.
- If you use a bootstrap or configuration server, set desktopHome in its response.
- Remove the deprecated keys coworkTabEnabled, chatTabEnabled and chatAdvancedFileAnalysisEnabled. While desktopHome is set they have no effect. Users see a warning from 17 November 2026 while one is still set. Anthropic accepts them until 12:00 PM PT on 17 December 2026; after that, releases that follow 4.4212 ignore them, while earlier releases go on reading them.
In the Enterprise Admin Console
- Sign in at claude.ai, switch to the organisation and open Organization settings.
- In the Desktop 3P section, open Capabilities and find Allowed surfaces.
- Turn on Opt into the new Chat and Cowork unified view. This sets the desktopHome key; turning it off removes the key, which means the 10 November default applies.
- To keep Chat and Cowork off, leave Chat and Cowork unified view switched off, which saves the value off. Switch it on to save standard instead.
- Click Save changes. Groups set on the Permission policies page have the same two switches in their Capabilities section, and Anthropic warns that from 10 November a group whose policy turns Cowork off gets Cowork unless these switches are set.
Retention settings to check at the same time
If you delete idle conversations automatically, Anthropic's 3 November changelog adds a trap: new conversations that are Cowork tasks are deleted by coworkSessionRetentionDays, not chatSessionRetentionDays. If you choose standard and only set the chat period, Cowork tasks are kept until their own period says otherwise. Set both. Anthropic also notes that the 10 November default applies even where chatSessionRetentionDays is set, so a retention setting will not hold the old view in place.
Common mistakes and how to fix them
- Relying on coworkTabEnabled set to false. From 10 November it no longer keeps Cowork off when desktopHome is unset. Set desktopHome explicitly.
- Typing Standard or OFF. Unrecognised values read as off, which can switch chat off for everyone by accident. Use lowercase standard or off.
- Turning off the opt-in switch in the Admin Console. That removes the key, so the default of standard applies after 10 November. Keep the opt-in on and set the second switch.
- Testing on an old app version. Anthropic says current versions apply the new default. Earlier releases go on reading the deprecated keys, so test on the version your users will run.
- Forgetting to tell people. If you turn Cowork on, explain what it can do and what it should not touch. Our guide to using AI at work safely and our template for an AI policy help.
Should you turn Cowork off at all?
We think most organisations should not switch it off permanently. Cowork is where Claude does its most valuable work, from building a spreadsheet out of a folder of invoices to drafting a report from research, and Anthropic gives admins finer controls than a single off switch: cloud sessions, the built-in browser, connector approvals and, on Enterprise, groups and custom roles. A short, deliberate pause while you review security settings is reasonable; Anthropic itself calls the prompt injection risk in Cowork non-zero. A permanent off switch mostly pushes people towards tools you do not manage. For the wider picture, see our Claude guide, how to use AI agents and how to roll out AI to your team.
Prompts to copy
Audit your Claude Desktop configuration
Here is our Claude Desktop managed configuration: [paste profile or registry export with secrets removed]. List every key that Anthropic has deprecated in favour of desktopHome, whether desktopHome is set and to what, and whether chatSessionRetentionDays and coworkSessionRetentionDays are both set. Flag anything whose value is not lowercase.
You get: A checklist of what to change before 10 November
Announce the change to staff
Write a short internal message for [company] staff explaining that from [date] Claude will [keep Cowork off / show chat and Cowork in one view]. Explain in plain English what Cowork can do, what they must not use it for under our policy ([paste key rules]), and who to contact with questions. Under 200 words.
You get: A clear staff notice that matches your decision
Decide between standard and off
We are deciding whether to set Claude Desktop's desktopHome to standard or off. Our constraints: [list security, compliance, data and team needs]. Give the strongest case for each option, the risks of each, and which extra Cowork controls we should review if we choose standard. Do not assume any setting exists unless I name it.
You get: A balanced decision memo for your security review