Google Rolls Out Gemini Computer Use for Mac Amid Apple Permission Changes
-
The Tech TL;DR:
- Gemini now operates Mac apps, files, and web tasks in an isolated background window with live preview, requiring Apple Silicon and macOS Sequoia 15.0 or later.
- Access is gated behind Google AI Plus, Pro, or Ultra subscriptions starting at $4.99 per month, initially for U.S. users aged 18 and older.
- Apple’s new split permission model for Device Control and Data Access may constrain how broadly the agent can operate across the system.
The rollout arrives in Gemini app version 27 Golden Gate and marks Google’s entry into desktop agentic control — a space where Meta, OpenAI, and Anthropic already have competing offerings.
Users invoke the capability by including "@Computer use" in prompts, directing the model to move through menus, fill forms, and move data across applications without seizing the primary cursor or display.
“Google hasn’t explained how Gemini will operate under the new restrictions,” The New Stack said, referencing Apple’s updated permission architecture that separates System & App Protection Override from Accessibility prompts.
Google’s announcement emphasizes that the agent runs in a separate background window, leaving the main screen free for concurrent work, and creates temporary backup copies of every file it touches — deleted after 24 hours or upon task completion, TechRadar reported. Permissions are granular: users select which apps Gemini may access, approve critical actions before execution, and can pause or stop the agent at any point.
Live preview streams while automation runs isolated
The background execution environment isolates Gemini’s clicks, keystrokes, and navigation from the user’s active session. A live preview streams inside the chat interface, providing observability without interrupting workflow. Macworld’s hands-on testing revealed that earlier app versions (1.147.5.1526) enforced a “built-in security boundary in Gemini’s Mac sandbox” that blocked automation attempts in Pixelmator Pro, QuickTime Player, and Finder until the updated build propagated.
Google has not disclosed whether the consumer Mac feature shares the same execution architecture as the developer-facing computer-use API introduced with Gemini 2.5 in October 2025 and later integrated into Gemini 3.5 Flash, The New Stack reported. The company also hasn’t detailed prompt-injection safeguards specific to the Mac app, though such protections are documented for the API.
Permissions After macOS 27.3
Apple’s beta changes split automation consent into two categories: System & App Protection Override and Accessibility. Apps requesting broad system access face explicit per-request prompts, and users can restrict or revoke permissions while an agent runs, The New Stack reported. Google’s announcement acknowledges the new controls but has not specified how Gemini’s background window maps to these categories or whether least-privilege access patterns are enforced at runtime.
With computer use in the Gemini app for Mac, you can now hand off tasks across your folders, apps, and the web, while your screen and cursor stay yours. Watch Gemini click, scroll, and type in a live background window, turning a quick sketch into a rigged 3D model in Blender.
Apple Silicon Macs required for computer use access
| Tier | Monthly Cost | Computer Use Access | Region |
|---|---|---|---|
| Google AI Plus | $4.99 | Yes | U.S. only (initial) |
| Google AI Pro | Higher | Yes | U.S. only (initial) |
| Google AI Ultra | Highest | Yes | U.S. only (initial) |
| Free | $0 | No | N/A |
Hardware requirements mandate Apple Silicon — M-series chips across MacBook Air, MacBook Pro, Mac mini, and iMac — running macOS Sequoia 15.0 or later, Android Authority reported. Intel Macs are excluded. The feature operates in English only during this initial phase.
Competitive Positioning
Google’s entry follows Anthropic’s Computer Use API (public beta, October 2024), OpenAI’s Operator research preview (January 2025), and Meta’s agentic tooling for desktop environments. Unlike the developer APIs from Anthropic and OpenAI, Gemini’s Mac implementation is a consumer-app feature, not a programmable interface — third-party developers cannot yet build atop the background environment, The New Stack confirmed.

Google demonstrated the agent converting a child’s butterfly sketch into a rigged 3D model in Blender, then exporting to an augmented-reality web app and a 3D-printable ring design, TechRadar reported. Practical workflows cited include batch photo editing, extracting structured data from local PDFs into shipping quotes, and compiling regional dashboard screenshots into presentations.
Google was founded in 1998 by Larry Page and Sergey Brin, who still control 56% of voting power through super-voting stock. The company reorganized under Alphabet Inc. in 2015, with Sundar Pichai becoming CEO of both Google and Alphabet by 2019. Google Search and YouTube remain the world’s two most-visited websites, and advertising generated 73% of revenue in 2025.
# Invocation pattern in Gemini for Mac
@Computer use "Extract weight, dimensions, and destination from PDFs in ~/Documents/Shipping and populate the carrier quote template in Sheets"
The feature rolls out immediately to eligible subscribers via gemini.google/mac. Google has not published a timeline for international expansion, additional language support, or a developer API for the Mac background environment.