Introducing computer use in Gemini 3.5 Flash

Gemini 3.5 Flash now includes built-in computer use capabilities, allowing the model to interact directly with desktop, mobile, and browser environments. By integrating this tool natively into the main model, developers can build agents that see, reason, and perform actions across various software platforms. This advancement is designed to improve performance for complex, long-horizon tasks such as continuous software testing and automated knowledge work.

To address security concerns, Google has implemented targeted adversarial training to mitigate risks like prompt injection. Enterprises can also utilize two optional safeguard systems that require human confirmation for sensitive actions and automatically halt tasks if potential security threats are detected. These features are intended to be used alongside standard security practices like sandboxing and strict access controls.

Developers and enterprises can access these capabilities through the Gemini API and the Gemini Enterprise Agent Platform. The technology is already being applied to tasks such as auditing documentation for accessibility and categorizing software features. By enabling agents to operate across professional applications, this update aims to streamline enterprise automation and increase the reliability of AI-driven workflows.