Anthropic opens Claude computer use to Team and Enterprise plans
Anthropic released computer use in beta for Claude Team and Enterprise plans on November 20, 2026.

· Last verified: 22 Nov 2026 (Anthropic)

Summary
- Anthropic released computer use in beta for Claude Team and Enterprise plans on November 20, 2026.
- The feature lets Claude see a user's screen and act inside approved desktop applications.
- Anthropic says there is no sandbox between Claude and a user's applications.
The latest
Claude can now see an employee's screen, click and type inside approved desktop applications, under a beta feature Anthropic opened to its Team and Enterprise plans on November 20, 2026. The company said the rollout is gradual and the capability ships disabled by default. It runs in Claude Desktop, and from web and mobile when computer access is switched on.
Details
- The mechanics: According to Anthropic's setup documentation, computer use works only in the new Claude experience where chat and Cowork are unified. Employees run it from Claude Desktop on macOS 14 or later and on Windows, or from web and mobile if computer access is enabled and a connected desktop sits open.
- Where it does not run: Anthropic said the feature does not work in tasks started through the separate Cowork option, nor in Claude Code. Organizations configured as HIPAA-ready, or using zero data retention for Claude Code, customer-managed encryption keys, or FERPA and K-12 settings, cannot enable it at all.
- Who controls it: Only an owner or primary owner can change the setting, through organization settings and then capabilities, under Claude Desktop. Anthropic said the change reaches user devices in roughly an hour, and sometimes longer. Computer use and computer access remain two separate settings.
- Roles and exceptions: On Enterprise, the user, admin, owner and primary owner roles receive the feature the moment it is switched on, and excluding someone requires moving them to a custom role and waiting 15 minutes. Team plans have no custom roles, so the setting applies to every user.
- Application lists: Administrators choose between blocking named applications or allowing only specified ones, with per-application user approval still required. Macs identify apps by bundle ID such as com.microsoft.Excel, Windows by executable name such as excel.exe or AUMID. Anthropic said typed identifiers are not validated.
- Device management: The MDM policy key computerUseEnabled set to false disables the feature, requires Claude Desktop 5.1231.0 or newer, and applies per device rather than per user. Anthropic said the key does not cover skill registration, Claude in Chrome, the built-in browser, Claude Code in the terminal, local MCP servers or desktop extensions.
- The stated risk: Anthropic's documentation states plainly: "Computer use has no sandbox between Claude and a user's applications." The company warned that on-screen content, including web pages and messages, may trigger unintended actions.
- Beyond folder limits: Anthropic said the feature is not confined to connected folders. Through approved applications, Claude can open, edit and delete files outside them, and read-only settings do not apply. Block and allow lists also miss Claude in Chrome, Claude Code, browser extensions, remote desktops, virtual machines and text copied from a blocked app.
- Screenshots and data: Screenshots are saved with the conversation and may contain personal or confidential data, according to Anthropic. An automated review process inspects them to check actions taken, and the company said it may retain a copy.
- Permissions required: On macOS, Claude needs accessibility and screen recording permissions, which administrators can grant in advance through a PPPC profile. Windows requires no system-level permission, Anthropic said.
Background
Anthropic first launched computer use on March 23, 2026 as a research preview inside Cowork, limited to the Pro and Max consumer plans. The November release is its first extension to business customers.
Between the lines
The absence of a sandbox, combined with Claude's ability to reach files outside connected folders through approved apps, shifts the security perimeter onto IT departments. The controls Anthropic provides are lists and roles, not isolation, and unvalidated app identifiers mean a typo in an allow list silently widens or narrows access. Gradual rollout and default-off suggest the company expects configuration work before activation.
What's next
Watch when the gradual rollout reaches individual organizations, whether Anthropic extends MDM coverage to Claude in Chrome and Claude Code, and whether the feature moves out of beta.
Source: Anthropic