Claude computer use hits Team and Enterprise: set the app list first
Claude computer use is now in beta on Team and Enterprise plans. It is off by default, has no sandbox, and admins get app allow and block lists. Set them first.
Claude can now drive the desktop apps on your team's computers. On November 20, Anthropic put Claude computer use in beta for Team and Enterprise plans. Claude sees the screen and works in the apps a user allows, like Excel or the old inventory tool with no API. Useful. It also puts an admin decision between your staff and an agent with mouse and keyboard access.
What actually happened
From Anthropic's admin setup guide:
- Off by default. Only an Owner or Primary Owner can turn it on, under Organization settings > Capabilities > Claude Desktop. The rollout to organizations is gradual.
- Where it runs. Claude Desktop on macOS 14+ and Windows, in the new unified Chat and Cowork experience. Users on web and mobile can reach their computer if the org's Computer access setting is on and Claude Desktop is open. It does not work in Claude Code.
- Two gates per user. Each user turns on "Enable computer use" on each computer, and still approves each app.
- App lists. Admins pick "Block these apps" or "Allow only these apps." Mac apps go in by bundle ID, Windows apps by file name. Changes can take about an hour to land.
- No sandbox. Anthropic says on-screen content like web pages or emails "could cause Claude to take unintended actions." Screenshots are saved with the chat.
- Folder limits don't apply. Read-only folder settings don't stop computer use. An approved app can change or delete files anywhere it can reach.
- MDM off switch. Set
computerUseEnabledto false at the machine level. MDM can turn it off, not on. - Excluded setups. HIPAA-ready, Zero Data Retention for Claude Code, customer-managed keys, and FERPA/K-12 orgs can't turn it on.
Why it matters for your business
Most small businesses run at least one tool with no API. Computer use is how an agent gets in. Do it on purpose.
The block list has gaps, and Anthropic lists them: websites in an approved browser, apps inside remote desktops, text copied from a blocked app, and Windows notifications that show up in screenshots.
Our order of operations:
- Start with "Allow only these apps." Name the two or three apps the job needs. Everything else is denied.
- Keep the browser off the list unless the task needs it. The browser is where prompt injection comes from.
- Pilot on one machine with one user and one task before you turn it on org-wide.
- Treat screenshots as records. If the screen shows customer data, the chat now holds it too.
Key takeaways
- Claude computer use is in beta on Team and Enterprise as of November 20
- Off by default; only Owners turn it on, and users approve each app
- No sandbox, and screenshots are saved with the chat
- Allow and block lists don't cover browsers' sites, remote desktops or notifications
- Use an allow list and pilot one job on one machine first
Got a back-office tool with no API? We scope computer-use agents to one job and the minimum app list, and we wire in the approvals before anything runs. See how we build guarded automations or tell us which screen eats your week.
Sources: Claude release notes, Anthropic: Set up computer use for Team and Enterprise plans.
- #claude
- #computer-use
- #anthropic
- #ai-agents
- #agent-governance
Tommy Rush — Founder, Rush Commerce
Operator turned builder. 15+ years running operations — now shipping the systems businesses run on. More
Get The Rush Report weekly — one email, zero fluff.
Keep reading
Together AI prompt leak: an AI coding agent committed the key
Together AI says an AI coding agent pushed a credential to GitHub, exposing stored prompt samples from June and July. ZDR is now default for new accounts only.
Read itSuperhuman Go AI assistant: proactive agents, Docs on Dec 1
Superhuman Go becomes a proactive AI assistant across mail, docs and 100+ tools, with triggered Docs agents in beta Dec 1. Scope what it can see first.
Read it