Google DeepMind rolled out computer use as a native capability in Gemini 3.5 Flash, merging the former standalone Gemini 2.5 model into the main Flash engine. The upgrade lets developers create agents that see, reason and act across browsers, mobile and desktop apps without external wrappers, and reduces latency of multi‑step workflows, cutting cycle time.

Gemini 3.5 Flash already excels at function calling, Search and Maps integration; adding built‑in computer use expands its reach into workflows such as testing and knowledge work across suites. Access comes via the Gemini API and the Gemini Enterprise Agent Platform, where developers can invoke the feature to audit apps, categorize UI elements or flag accessibility gaps.

To curb prompt‑injection threats, DeepMind applied targeted adversarial training and released two optional safeguard layers: mandatory user confirmation for sensitive actions and automatic task termination on detected injection attempts. The company advises coupling these controls with sandboxing, human‑in‑the‑loop checks and strict access policies. These measures have already prevented several high‑risk exploits in pilot deployments, and early customers report measurable productivity gains from the integrated computer‑use capability.