Google’s Gemini can now operate your Mac. The company announced computer use for the Gemini app on Mac on November 19, a feature that lets the AI assistant click, type and fill in forms across desktop apps, local files and websites to finish multi step tasks on your behalf. You describe the job, Gemini does the clicking, and you watch it happen in a live preview while your own screen and cursor stay free.
It is the biggest expansion of Gemini on the Mac since the native app launched in April, and it puts Google ahead of Apple on Apple’s own platform, at least for now. But Google’s own help pages are unusually blunt about the risks, and the feature arrives three days after Apple added new permission controls for exactly this kind of software in the macOS 27.3 developer beta.
What computer use does on a Mac
Computer use turns Gemini from a chatbot into an operator. Add @Computer use to a prompt and describe the task, and Gemini works through the steps itself: opening apps, navigating menus, typing into fields, moving files, and filling in forms. Google’s examples include modeling and rigging a sketch in Blender, a 3D app that now appears in the Gemini app’s Customize and Connectors menu alongside Autodesk Fusion.
Until now, the Mac app’s Tasks feature, formerly called Gemini Spark, could work with local files and connected Google services like Gmail and Google Calendar, but it could not operate other desktop software. Computer use fills that gap. Gemini works inside an isolated background window with a live preview in your chat, so you can keep using your Mac while it works on something else.
The controls are designed to keep a human in the loop. You choose which apps Gemini can access, you can pause a task at any time, and Gemini asks for approval before key actions. It pauses for logins and payments, and asks for confirmation before sensitive actions. The first time Gemini wants to use an app, it asks whether to deny access, allow it once, or always allow it.
Who gets it, and what it costs
Computer use is rolling out in English in the United States to personal Google accounts for users aged 18 and over. It requires a Mac with Apple Silicon running macOS Sequoia 15.0 or later, with at least 8GB of RAM. Intel Macs are excluded, and Google has not said when the feature will reach other languages or countries, or whether it will come to free Gemini accounts.
Google’s announcement lists AI Plus, Pro and Ultra subscribers, but its support documentation currently specifies only Pro and Ultra for controlling desktop apps, so Plus subscribers should check before assuming they have access. The plans start at $4.99 a month for AI Plus, $19.99 for AI Pro, and $99.99 or $199.99 for AI Ultra. The Gemini app for Mac is a free download from gemini.google/mac, and the feature is rolling out gradually, so not every eligible account will see it on day one.
| Plan | Monthly price | Computer use |
|---|---|---|
| Google AI Plus | $4.99 | Listed in announcement, unclear in support docs |
| Google AI Pro | $19.99 | Yes |
| Google AI Ultra | $99.99 / $199.99 | Yes |
For Indian readers, $4.99 works out to roughly 440 rupees a month at current exchange rates, though Google’s India pricing may differ when the feature arrives there. No India launch date has been announced.
Google’s own warnings are worth reading first
Handing an AI agent control of your computer is a bigger step than chatting with one, and Google’s help pages say so plainly. Before Gemini can operate Mac apps, you must grant macOS Accessibility and Screen and System Audio Recording permissions. From there, the risks Google lists include prompt injection, where malicious instructions hidden in ordinary content on screen redirect Gemini into actions in another app it has permission to use, and access to active sessions in apps that are already signed in, including two factor authentication codes visible in desktop or messaging apps.
Granting Gemini access to code editors or developer tools carries a further warning: those permissions could enable shell command execution or broader changes to the file system, beyond the files involved in the original task. Gemini can attempt temporary backups when its on demand backup setting is on, but Google warns that some apps do not support backups and certain changes cannot be undone.
That caution is well timed. A day before Google’s announcement, Reuters reported that Meta’s Muse agent used Mac accessibility and screenshot tools to read Signal messages in tests. Nothing in the reporting suggests Gemini has done the same, but the two features share the same permissions, and the same trust problem.
Apple appears to have seen this coming. The company introduced new permission controls in the macOS 27.3 developer beta 2 on November 16, three days before Google’s announcement, letting users limit access to specific apps and folders and configure apps to ask permission each time. Users who install apps asking for Mac permissions now will not have to re-grant those permissions after macOS 27.3 ships.
What to do if you try it
The safe way to start is narrow. Allow Gemini access to one or two low stakes apps, like a notes app or a media converter, and choose allow once rather than always allow. Keep code editors, terminals, password managers, messaging apps and banking apps off the allowed list entirely. Watch the live preview instead of letting tasks run unattended, and check the result before Gemini moves on to the next step.
Computer use is the direction every AI assistant is heading. Anthropic’s Claude, OpenAI’s ChatGPT agent and now Google’s Gemini all want to stop answering questions and start doing the work. Google is the first to ship it as a consumer feature on the Mac, and the underlying model, Gemini 2.5 Computer Use, has been available to developers since October 2025. Google has not said whether the feature will come to the Windows version of the Gemini app.
The convenience is real, and so is the risk. An agent that can click anything can click the wrong thing, and an agent that can read your screen can read the parts you would rather it did not. Try it on tasks you can check in seconds, keep the sensitive apps locked out, and treat always allow as a decision you make per app, not a default.
For more on Google’s AI lineup, see our coverage of Gemini 4 Neon, the new mid tier coding model and Gemini 4 Hydrogen, Google’s cheapest Gemini 4 model. For the other side of Mac AI agents, read Perplexity Teammates on the Mac.
