Google adds native computer‑use capability to Gemini AI, expanding it across Android, Chrome, Workspace and more

Google has launched Gemini 3.5 Flash with a built‑in “computer use” feature that lets the model see, reason about and interact with on‑screen elements across browsers, desktop applications and mobile devices. The capability, previously offered only as a separate Gemini 2.5 Computer Use model, is now native to the main Gemini Flash model, scoring 78.4 % on the OSWorld‑Verified benchmark. Google introduced safety measures including targeted adversarial training, optional enterprise safeguards that require explicit user confirmation for sensitive actions and automatic termination of tasks upon detecting prompt‑injection attempts.

The update is being rolled out throughout Google’s ecosystem. Android 16 ships with Gemini as the default assistant, while a prototype brings Gemini to signed‑out and Incognito Chrome sessions. Gemini functions are embedded in Google Workspace for text generation, summarisation and code‑snippets. The model also appears in Android Automotive vehicles such as Volvo and Polestar models, and has been added as a coding assistant in Apple’s Xcode IDE. Early‑stage Mac features, including Gemini Live voice chat and Magic Pointer, are being tested.

Developers can access the new functionality via the Gemini API and the Gemini Enterprise Agent Platform, with demo environments hosted on Browserbase. Google emphasises the potential for enterprise automation, continuous software testing and long‑horizon knowledge‑workflows, while urging organisations to adopt a defence‑in‑depth strategy for secure deployment.