Google launched computer use in the Gemini app for Mac on November 19, 2026. In a blog post, Michael Friedman, Group Product Manager for the Gemini app, said the feature “helps you get multi-step tasks done across your local files, desktop apps, and the web.” Google said it is rolling out in English in the U.S. for personal Google accounts, age 18 or older, on Google AI Plus, Pro and Ultra, and that it requires an Apple Silicon Mac running macOS Sequoia 15.0 or later.

#What changed?

Google said users describe what they need, and Gemini handles the clicks, typing, menus and forms. “Just describe what you need, and Gemini handles the clicking, typing, menus, and forms for you.” Google’s examples include color-grading photos or setting up 3D scenes, filling shipping quotes from local PDFs, building a Google Slides deck from live dashboards, and staging resale listings from a local photo folder.

Google said: “Gemini works inside an isolated background window with a live preview right in your chat.” The help page says Gemini “moves the work into a virtual screen so you can continue working in other apps.”

On privacy and control, the blog says users “choose which apps Gemini can access, watch every step live, pause at any time, and approve key actions before they happen.” Google’s computer-use help page says the first time Gemini works with an app, it asks whether to deny, allow or always allow permission. “It can only control apps that you grant permission for Gemini to access.” If the user chooses “always allow,” Gemini will not ask again unless the app is removed from the Allowed apps list. “For sensitive tasks, Gemini will ask you for confirmation before continuing.”

Google’s tasks help pages, which the computer-use page links to, say “very sensitive actions require your permission each time.” Those pages list sending communications, deleting data, sharing sensitive info, making purchases, completing financial transactions, filling out and submitting web forms, and signing in with Google or Google Password Manager. They also say Gemini may pause and ask the user to “take control” to enter passwords or payment details. A related permissions page for tasks says Gemini is designed to ask before it deletes data, sends messages or emails in Google apps, or handles payments or other sensitive third-party tasks.

The computer-use help page says Gemini will try to create a temporary backup while working, but is not always able to do so. Backups are kept for 24 hours, or until the next prompt, and some app content may not be backed up. The same page warns, in Google’s wording, that Gemini can make mistakes, share info with desktop apps, and reach logged-in accounts and active sessions. It says some actions are irreversible, that untrusted content on screen in one app can manipulate Gemini into taking actions in other allowed Mac apps, that permission for certain local apps such as code or text editors or developer tools can enable shell execution or broader file system changes, and that Gemini can read or interact with multi-factor authentication prompts or codes visible in desktop or messaging apps. Automating certain third-party desktop apps may conflict with those apps’ terms of service, the page says. The tasks help page defines prompt injection as “an attempt to elicit an unintended or harmful response from AI tools like Gemini.”

#Who can use it and when?

Computer use is rolling out in English in the U.S. in the Gemini app for Mac, as of Google’s November 19, 2026 post. A footnote on that post says: “Requires an Apple Silicon Mac running macOS Sequoia 15.0+. Available for personal Google accounts 18+, available to Google AI Plus, Pro and Ultra users.” Google’s computer-use help page says the feature “is only available in Gemini tasks for personal accounts, not work or school accounts.” The Gemini app for Mac itself can be used with a work or school account that an administrator has enabled, so that exclusion is specific to computer use. The general Mac app page also says installation needs 8 GB of RAM or more and at least 200 MB of available disk space, and that some features may not work if the app is out of date.

Computer use runs inside Gemini tasks, and still requires a Google AI Plus, Pro or Ultra plan in English in the U.S. That is narrower than tasks in Gemini becoming free in six countries. Google’s computer-use help page says that for better quality responses, users should choose Gemini 4 Hydrogen or newer models. See the Gemini 4 Hydrogen launch.

#How to get it

Google said: “Download the app at gemini.google/mac to get started.” (gemini.google/mac) The computer-use help page lists these steps:

  1. Sign in with a personal Google account that is 18 or older and on Google AI Plus, Pro or Ultra.
  2. Turn on Computer use, and turn on macOS Accessibility and Screen & System Audio Recording if they are not already on.
  3. Below the chat box, click Task, mention @Computer use and explain what Gemini should do.

The blog says users can pause at any time. The computer-use help page says: “To end a task at any time, click Stop task.”

#How does it compare?

Computer use in the Gemini app for Mac arrives after other desktop agents, and only for paid personal accounts in the U.S. TechCrunch wrote that Google “is a little late to the game with this launch,” pointing to agentic features already offered in desktop apps by OpenAI, Anthropic and Meta. On Apple silicon Macs, Perplexity Computer can already run fully local. OpenAI has also shipped a cheaper model aimed at computer-use work, GPT-6.1 Terra.

The Verge reported that Gemini is introducing the feature ahead of Apple launching new restrictions for apps wanting “full disk access” on Mac, citing potential risks from AI agents with computer use. Google’s own pages name Accessibility and Screen & System Audio Recording as the macOS permissions computer use needs. See the site’s coverage of macOS 27.3 beta 2’s finer disk-access and app-control settings and the guide to controlling AI coding agent network access on a Mac.

Our analysis: Google is putting a computer-use agent into a mainstream consumer Mac app, but only for paid personal accounts in the U.S., and only inside Gemini tasks. Users pick the apps Gemini may use, can watch a live preview, and confirm sensitive or first-time actions, rather than being asked before each click. That makes the help-page risk list, especially logged-in sessions and on-screen authentication codes, the part readers should weigh before leaving an app on always allow.

#What is still unclear?

What do Google AI Plus, Pro and Ultra cost, and does computer use have its own usage limits? Which Gemini app for Mac version is required? When might computer use expand to other countries, languages, Windows, Intel Macs, or work and school accounts? How is computer-use preview data stored, and is it used for model training? Which computer-use steps, beyond the examples on Google’s tasks help pages, count as sensitive enough to need confirmation? How will Apple’s macOS 27.3 agent-control changes interact with Gemini’s computer use?