Google has integrated its AI model, Gemini 3.5 Flash, to directly control computers, browsers, and mobile devices. This allows developers to build agents that can work across various environments, such as software testing or office automation. Gemini 3.5 Flash scores 78.4 on the OSWorld benchmark, beating its predecessor and some other models. To prevent prompt injection attacks, Google uses adversarial training and enterprise safeguards. This feature is available through the Gemini API and the Gemini Enterprise Agent Platform.