Google Launches Native Computer Use in Gemini 3.5 Flash for Agentic AI Tasks
Summary
- • Computer use is now built natively into Gemini 3.5 Flash, replacing the need for a separate model.
- • Developers can build agents that control browsers, mobile, and desktop environments.
- • Google added adversarial training and enterprise safeguards against prompt injection attacks.
- • Available now via the Gemini API and Gemini Enterprise Agent Platform.
Details
Computer use native in 3.5 Flash
Computer use is now a built-in tool in Gemini 3.5 Flash — previously only available as a standalone Gemini 2.5 model, this integration dramatically expands developer access.
Browser, mobile, desktop control
Agents built with 3.5 Flash can see, reason, and take action across browser, mobile, and desktop environments, enabling full UI automation.
Long-horizon enterprise use cases
Google targets continuous software testing and knowledge work across professional applications as primary enterprise use cases for the capability.
Adversarial training for prompt injection
Google applied targeted adversarial training to mitigate prompt injection risks — a key threat for agents operating in live, untrusted environments.
Two optional enterprise safeguards
Enterprises can require explicit user confirmation for sensitive or irreversible actions, and/or enable auto-stopping of tasks when indirect prompt injection is detected.
Available via API and Enterprise Platform
Developers can access the feature through the Gemini API and Gemini Enterprise Agent Platform, with a demo environment hosted by Browserbase.
Details of Google's native computer use integration in Gemini 3.5 Flash and its safety architecture.
What This Means
By baking computer use directly into Gemini 3.5 Flash rather than offering it as a niche separate product, Google is making agentic desktop and browser automation mainstream for enterprise developers. This move directly challenges Anthropic's Claude computer use offering and accelerates the industry shift toward AI agents that can autonomously operate software interfaces. The accompanying safety measures — particularly prompt injection adversarial training and enterprise safeguards — signal that Google is taking the security risks of autonomous agents seriously as the technology moves into production. For enterprises, this lowers the barrier to building custom automation agents that can navigate real-world software environments without requiring bespoke integrations.