Claude Computer Use on Team and Enterprise: What Businesses Should Know
On 20 November 2026 Anthropic made computer use available in beta on Claude Team and Enterprise plans. Claude can look at a user's screen and work inside the desktop apps that person allows. It is off by default, and only the organisation owner can turn it on. Here is what that means for a Malaysian company that runs its work on desktop software.
What changed
Anthropic's release notes for 20 November 2026 add computer use to the Team and Enterprise plans as a beta that is rolling out to organisations gradually. Computer use lets Claude take screenshots of a user's screen and then click, type and move between the desktop apps that user approves. Until now this kind of screen control was mainly something developers set up themselves through the API. Now it sits inside the normal Claude Desktop app that office staff already use.
The setup article lists the details that matter for an IT person:
- It is off by default. Only an Owner or Primary Owner can switch it on, under Organization settings, Capabilities, after turning on the unified Chat and Cowork experience.
- It works in Claude Desktop on macOS 14 or later and on Windows. On a Mac it needs the Accessibility and Screen Recording permissions.
- Staff can also start a task from Claude on the web or on their phone, but only when the organisation allows computer access and a computer with Claude Desktop open is connected.
- Each user still has to turn on computer use in Claude Desktop, on each computer, and approve each app.
- On Enterprise, owners can keep it away from chosen roles. On Team, the setting covers every user, because Team has no custom roles.
- Owners can keep a block list or an allow list of apps, for example excel.exe on Windows or com.microsoft.Excel on a Mac.
The article does not state a region limit, so a Malaysian company on a Team or Enterprise plan should see it as the rollout reaches its organisation. Some organisations cannot turn it on at all, including those using the HIPAA-ready configuration, customer-managed encryption keys, or Zero Data Retention for Claude Code.
Why it matters for Malaysian companies
Many companies here still run part of their day on desktop software that has no API: an older accounting package, a stock program installed on one office PC, a supplier portal, or a spreadsheet that has grown into a small system. When I connect systems for clients, the cleanest route is always an API. When there is none, the work usually falls back on a person copying figures from one window into another. That is exactly the gap computer use is built for.
Some realistic first tasks:
- Reading a batch of purchase orders in one app and keying the line items into an older stock program that has no import function.
- Pulling month-end figures from several desktop reports into one spreadsheet so a manager can review them.
- Checking a list of records against what is shown in another program and noting the differences for a person to fix.
- Filling in the same form many times from a prepared list, with a person reviewing the result.
These are the repetitive, screen-to-screen jobs that take staff hours each week. They are also the jobs where a mistake is easy to spot when someone checks the output.
Limits and cautions
Anthropic is direct about the risks, and a company should read them before switching anything on.
- No sandbox. The setup article says there is no sandbox between Claude and a user's applications. Content shown on screen, such as text in an email or a web page, could lead Claude to take actions nobody intended.
- Wider file access. Computer use is not limited to the folders a user connected to Claude. Through an approved app, Claude can open, change or delete files outside them, and read only folder settings do not apply.
- Screenshots are stored. Screenshots are saved with the chat and can include personal or confidential information. The automated check that reviews Claude's actions can also see the latest screenshot. For a Malaysian company this is a PDPA question: decide which screens may hold customer or staff personal data before allowing them.
- App lists have gaps. The allow and block lists do not cover browsers driven through Claude in Chrome, websites, apps inside remote desktops or virtual machines, or text copied from a blocked app.
- It is a beta. It is slower than an API integration and it can make mistakes, so every result that goes into accounts, stock or customer records needs a person to check it.
How I would start
Start small and keep it narrow. Pick one task that is boring, repeated and easy to check. Use an allow list instead of a block list, so Claude can only touch the two or three apps that task needs. Run it on one staff computer, with a person watching the first runs, and keep the source data in front of you to compare. If the company is on Enterprise, keep it limited to the roles doing that task.
I would also treat computer use as a bridge, not the final design. If a task runs every day, it is usually worth asking whether the old program can export a file, or whether a small integration can read the data directly. That kind of link is faster and does not depend on what is on screen. Systems that keep records in one place, as I describe in what to prepare before building an ERP, remove much of the copying in the first place.
My own team already works this way in a different form. Our WhatsApp AI desk gives an AI agent tools that read our own systems directly, rather than reading a screen, so answers come back in seconds and every step is logged. Computer use fills the gap where no such connection exists yet.
If you want help deciding which tasks suit screen automation and which need a proper integration, see the work I take on or how we can work together.
Sources
- Release notes: Computer use (beta) for Team and Enterprise plans (Anthropic, 2026-11-20)
- Set up computer use for Team and Enterprise plans (Anthropic, 2026-11-20)
