Google Confirms Gemini Upgrade: AI Can Now Control Your Computer

Google just gave its Gemini AI the ability to see and control your computer screen — and it's putting the privacy burden on you. The company confirmed on June 27, 2026, that its new Gemini Computer Use feature lets the AI agent navigate your desktop, open applications, fill forms, and click buttons, all through a browser extension. But in a move that has alarmed privacy advocates, Google explicitly warned that users are solely responsible for any data the AI accesses, placing the onus squarely on individuals rather than the tech giant.

Google's announcement marks a significant leap in AI agent capabilities, following similar moves by OpenAI and Anthropic in 2025 and 2026. The feature is designed to automate repetitive tasks like filing reports, booking travel, or managing spreadsheets, positioning Gemini as a direct competitor to Microsoft's Copilot and other desktop AI assistants. However, the lack of built-in privacy safeguards has drawn sharp criticism from cybersecurity experts and consumer groups.

According to Google's official blog post, Gemini Computer Use works by streaming your screen to the AI, which then issues commands to control the mouse and keyboard via a Chrome extension. The system can handle multi-step workflows, such as logging into accounts, copying data between apps, and submitting web forms. Early testers reported that the AI could accurately complete tasks like ordering groceries or scheduling meetings with minimal errors, but also noted that it occasionally accessed unintended files or exposed sensitive information.

Named individuals include Google CEO Sundar Pichai, who called the feature 'a new era of personal productivity,' and Gemini product lead Claire Liu, who acknowledged the privacy trade-offs. The rollout begins next week in the US for Gemini Advanced subscribers, with EU and Asian markets following in August. Google also released a 30-page safety whitepaper that lists potential risks — from accidental data leaks to malicious exploitation — but offers no concrete technical protections beyond user-discretion warnings.

Analysts warn that the 'user responsibility' model is a dangerous abdication. 'This is like selling a self-driving car and telling the buyer they're liable for all crashes, even software failures,' said Dr. Elena Rossi, a Stanford cybersecurity researcher. The broader implication is that AI agents may outpace safety regulations, forcing regulators to scramble for oversight frameworks. Google's approach contrasts with Apple's, which has limited on-device AI to sandboxed environments.

The outlook is uncertain. While Gemini Computer Use could transform how millions interact with computers, privacy watchdogs are already calling for federal investigations. Google promises regular updates and an opt-in transparency dashboard, but critics say these measures fall short. The next milestone to watch is the EU's response under the AI Act, which could force Google to implement mandatory data filters. For now, users face a stark choice: productivity gains or data autonomy.