What happened
Google integrated "computer use" capabilities directly into Gemini 3.5 Flash, transitioning it from a standalone Gemini 2.5 model to a native feature within the main Flash model. This enables developers to build custom agents that perceive, reason, and act across browser, mobile, and desktop environments via the Gemini API and Gemini Enterprise Agent Platform. The update includes targeted adversarial training and two optional enterprise safeguard systems: requiring explicit user confirmation for sensitive actions and automatically stopping tasks if an indirect prompt injection is identified.
Why it matters
Platform engineers and security architects gain a unified model for agentic workflows, reducing complexity and enhancing reliability. The native integration of "computer use" into Gemini 3.5 Flash improves performance for long-horizon and enterprise automation tasks like continuous software testing. Optional safeguards offer mechanisms to control agent actions and detect prompt injection, addressing critical security constraints for enterprise deployments. This follows Google's recent expansions of Gemini capabilities, such as the Gemini 3.5 Live Translate.
