What was announced
On June 24, 2026, Google announced that computer use is now a built-in tool in Gemini 3.5 Flash. Developers can build agents that see the screen, reason and act (click, type, navigate) across browser, mobile and desktop environments.
Previously this capability only existed as a standalone Gemini 2.5 Computer Use model. Now it is integrated into the main Flash model, which we covered at launch in May. Google calls it its best performance yet on computer-use tasks, aimed at long-horizon automation such as continuous software testing and work across enterprise apps.
Where it's available
- Gemini API and Gemini Enterprise Agent Platform, for developers and enterprises. It is not a feature for everyday Gemini app users.
- A demo hosted by Browserbase is available.
Note: Google's "What's new in Gemini 3.5 Flash" docs page still said computer use was not supported, while the model card lists it as "Supported (Preview)". Treat it as a preview.
Safety
An agent that controls a computer can be tricked by prompt injection hidden in pages or documents. To reduce the risk, Google:
- used targeted adversarial training for computer use;
- released two optional enterprise safeguards: require user confirmation before sensitive or irreversible actions, and automatically stop the task when indirect prompt injection is detected.
The safeguards are opt-in. It is up to developers to turn them on.
KGA IT's recommendations
- Enable both safeguards in any real deployment.
- Run the agent in an isolated environment (sandbox or VM) with least-privilege accounts.
- Keep a human approving payments, outgoing emails and deletions.
- Start with repetitive, low-risk tasks such as UI testing and public data collection.
Sources
- Google, "Introducing computer use in Gemini 3.5 Flash" (2026-06-24): https://blog.google/innovation-and-ai/models-and-research/gemini-models/introducing-computer-use-gemini-3-5-flash/
- Android Authority (2026-06-25): https://www.androidauthority.com/gemini-3-5-flash-computer-use-model-3681381/
- Google AI for Developers, Gemini 3.5 Flash model page: https://ai.google.dev/gemini-api/docs/models/gemini-3.5-flash
Want to automate tasks with AI agents without opening security gaps? KGA IT offers a free consultation: kga-it.com.