Introducing computer use in Gemini 3.5 Flash

Open original

Google integrates its computer use tool into Gemini 3.5 Flash, allowing agents to see, reason and act across browser, mobile and desktop environments for tasks like continuous software testing and accessibility audits. It uses adversarial training plus optional safeguards—explicit confirmations and auto-stop triggers—to reduce prompt-injection risks, and is accessible via the Gemini API and Enterprise Agent Platform.

  • Gemini 3.5 Flash now has computer use built directly into the main model, merging what was previously a standalone Gemini 2.5 feature so the same model handles chat, function-calling, and UI automation across browser, mobile, and desktop.
  • Google used targeted adversarial training to curb prompt-injection attacks, plus optional safeguards—mandatory user confirmation before sensitive/irreversible actions and auto-stop on suspicious prompt patterns.
  • Available now via the Gemini API and Gemini Enterprise Agent Platform, with a live demo on Browserbase and reference code in the docs.

Gemini 3.5 Flash now includes a built-in computer use tool, bringing the standalone Gemini 2.5 computer use features directly into the main Flash model. That means the same model you call for chat and function-calling can now automate tasks across browsers, mobile apps and desktop environments. Google says this upgrade boosts performance on long-horizon workflows like continuous software testing and knowledge work in enterprise apps. To tap into the new feature, developers use either the Gemini API or the Gemini Enterprise Agent Platform. In practice, you can ask 3.5 Flash to scan an application’s UI and list out its menu items or audit your own docs for accessibility problems. Under the hood, Google applied targeted adversarial training to limit prompt-injection attacks when agents operate in live settings. On top of that, two enterprise safeguards are optional: one forces the agent to pause and ask for user approval before any sensitive or irreversible action; the other automatically stops a task if it spots a suspicious prompt pattern. Google recommends pairing these controls with sandboxing, human review steps and tight access management. If you want to try it yourself, there’s a demo on Browserbase and full reference code in the Gemini documentation.

Ask about this article