Google has unveiled Gemini 3.5 Flash, an updated version of its AI model that now includes native “computer use” capabilities. This enhancement allows developers to create intelligent agents capable of interacting directly with digital environments across browsers, mobile devices, and desktop systems.
Previously, such functionalities were confined to the standalone Gemini 2.5 model. By integrating these features into Gemini 3.5 Flash, Google aims to improve performance and simplify deployment for developers focusing on automation tools and enterprise workflows.
Enhanced Agentic Behavior
The latest update introduces advanced agentic behavior, enabling AI systems to “see, reason, and act” within computing environments. These agents can perform complex, multi-step tasks such as navigating web interfaces, conducting software testing, and managing enterprise applications. Additionally, Gemini 3.5 Flash builds upon existing strengths like function calling and integration with tools such as Search and Maps. With the new computer use capability, the model can now handle long-duration tasks that require persistent interaction and contextual decision-making.
Security Measures and Recommendations
Recognizing the potential cybersecurity risks associated with autonomous agents accessing sensitive systems, Google has implemented several safeguards. Targeted adversarial training has been introduced to mitigate prompt injection attacks, which can manipulate AI behavior through malicious instructions. Furthermore, two enterprise-grade protections have been added: mandatory user confirmation for sensitive or irreversible actions, and automatic task termination upon detection of indirect prompt injection attempts.
Google advises organizations to adopt a defense-in-depth strategy, combining these protections with sandboxing, strict access controls, and human-in-the-loop verification. Despite these measures, security experts caution that autonomous agents interacting with live environments could introduce new exploitation vectors if not properly configured.
Gemini 3.5 Flash with computer use capabilities is now accessible through the Gemini API and the Gemini Enterprise Agent Platform. Developers can also experiment with the feature in a live demo environment hosted by Browserbase. Early adopters are already utilizing the technology for automation-intensive tasks, including continuous software testing and enterprise knowledge workflows. To facilitate development and integration, Google has released reference implementations on GitHub.
The integration of built-in computer use in a mainstream AI model marks a significant milestone for both innovation and security. While the technology offers powerful automation possibilities, it also raises concerns about unauthorized actions, data exposure, and AI-driven exploitation. Organizations adopting agent-based AI must prioritize secure deployment practices and continuously monitor agent behavior to mitigate emerging threats.
