What shipped

Computer use lets Claude see a user's screen and work in the desktop apps that user allows. Anthropic describes it plainly: Claude clicks and types like you would, on Claude Desktop. The feature has existed for consumer plans as a research preview since March 2026. What changed on November 20, 2026 is the audience: the release notes move it into beta for Team and Enterprise plans, with a gradual rollout, which means some organizations will not see the setting yet.

That gradual part matters for expectations inside a company. If the toggle is missing, Anthropic says it either has not reached your organization or your organization is one that cannot have it. There is no date for full availability.

The controls are the story

The business version is built around an administrator saying yes on purpose. Computer use is off by default for every Team and Enterprise organization. On Enterprise, it sits inside custom roles, so an admin can give it to some roles and not others. Team plans have no custom roles, so the switch covers every user at once. Anyone with a built-in role, User, Admin, Owner or Primary Owner, gets it as soon as it is turned on. To keep it from a specific person on Enterprise, an admin has to move them to a custom role first, and Anthropic advises waiting 15 minutes before switching the feature on.

Admins also get two app lists, an allow list and a block list, and only one is in use at a time; only an Owner or Primary Owner can change them, and users still approve each app themselves. Claude cannot click, type or appear in screenshots inside a blocked app. That sounds complete until the setup article lists what the lists do not cover, which includes Claude in Chrome, the built-in browser, Claude Code, websites and browser extensions, other accounts, and apps inside a remote desktop or virtual machine window. The boundary is the native app, not the work happening inside a browser window.

The boundary is wider than the lists suggest in another way. Anthropic states directly that computer use has no sandbox between Claude and a user's applications. The folder limits an organization sets for connected folders do not apply to it, and neither do "Read only" markings: through the apps a user approves, Claude can open, change or delete files outside those folders.

An hour or more to switch off

The detail most likely to surprise an IT team is the timing. Turning computer use off at the organization level can take about an hour, sometimes longer, to reach each user's Claude Desktop. Until it lands, Claude can still see the screen and use the apps the user allowed. The same delay applies when an app is newly blocked, when lists are switched, and when a role loses access: until then, Claude can still use an app that has just been denied.

There is a harder control at the device level. An MDM policy key, computerUseEnabled set to false, keeps computer use off on a chosen computer whichever account is signed in. The key needs Claude Desktop 5.1231.0 or later, and it is read only when the app starts, so it takes effect after Claude Desktop is fully quit and reopened; closing the window is not enough. An older version does not know the key exists, so it does not turn computer use off. Anthropic is also direct about the key's limits: it does not cover Claude in Chrome, the built-in browser, Claude Code in a terminal, local MCP servers or desktop extensions, and it is not a lock against someone who can change the app or the machine's settings. MDM cannot do the rest either: it cannot turn computer use on, cannot switch on a user's own per-computer setting, and cannot approve apps for users.

What Claude sees, and keeps

Computer use runs on screenshots. They are saved with the chat, and Anthropic notes they can include personal or confidential information. The automated check that reviews Claude's actions can see the latest screenshot, and Anthropic can keep a copy of it in its record of that check. There is no separate retention promise in the setup article beyond the Commercial Terms and Data Processing Addendum.

Whole categories of organization are excluded outright. HIPAA-ready configurations, Zero Data Retention for Claude Code, customer-managed encryption keys and FERPA or K-12 settings cannot enable the feature. If your compliance posture depends on any of those, this beta is not a decision to make. It has already been made.

The practical read

For a small Team-plan organization, the choice is blunt: everyone gets screen-driving Claude, or no one does, and screenshots of real work sit inside chat history. For an Enterprise admin, the work is in the sequencing. Set roles before enabling, build the block list before the first user asks, push the MDM key to machines that should never have it, and update those machines past Claude Desktop 5.1231.0 first, because on an older build the key is not read at all.

Sources

  1. [1] Anthropic, Claude release notes, November 20, 2026Read source
  2. [2] Anthropic, Set up computer use for Team and Enterprise plansRead source