Google DeepMind integrates computer-use tooling into Gemini 3.5 Flash

Google DeepMind announced that computer-use capabilities are now a built-in tool in the Gemini 3.5 Flash model, enabling developers to build agents that can see, reason, and take action across browser, mobile, and desktop environments. Previously, this capability was available only as a standalone model (Gemini 2.5 computer use).

The integration is positioned as delivering improved performance for long-horizon and enterprise automation tasks, such as continuous software testing and knowledge work across professional applications. Developers can access the feature via the Gemini API and the Gemini Enterprise Agent Platform.

To address prompt-injection risks in live environments, Google DeepMind implemented targeted adversarial training for computer-use in Gemini 3.5 Flash. The company also introduced two optional enterprise safeguard systems: one that requires explicit user confirmation for sensitive or irreversible actions, and another that automatically stops tasks if an indirect prompt injection is detected.

Google DeepMind recommends combining these safeguards with secure sandboxing, human-in-the-loop verification, and strict access controls as part of a defense-in-depth approach. Additional safety measures are documented in the company’s best practices guidance.

The announcement includes examples of computer use in practice, such as analyzing the Gemini app to return a categorized list of features and auditing documentation for accessibility issues.