Google Cloud Innovates with AI Threat Defense Against Cyber Threats

Google Cloud recently unveiled a new platform called "AI Threat Defense," designed to automate the detection, assessment, and remediation of security vulnerabilities in enterprise systems. This initiative leverages a set of technologies, some of which were acquired by Google, to provide an integrated and responsive solution to the growing cyber threats.

The platform consists of four main components: Gemini, which handles code analysis, Wiz, which assesses risks in cloud infrastructures, Codemender from Deepmind, which generates patches, and Mandiant, which brings its expertise in cyberattacks. Wiz, acquired by Google in March 2025 for $32 billion, plays a crucial role in identifying vulnerable systems, such as exposed servers, APIs, or credentials within the enterprise IT environment. An agent then simulates which vulnerabilities can actually be exploited.

To ensure comprehensive coverage, Google employs multiple AI models simultaneously. The performance of these models varies significantly depending on the task: some are better for application logic, others for cloud configuration or binary analysis. No single model can detect all vulnerabilities. Less expensive models are used for continuous checks, while advanced models focus on critical systems.

Codemender, introduced by Deepmind last fall, operates directly within the development environment to replace vulnerable code and modernize legacy codebases into modern, memory-safe languages. Before deploying a patch, the system automatically generates tests to verify each patch, thus ensuring traceability of their origin.

During live operations, agents from Google Security Operations take over to track active attacks. This automated approach has become essential, as new AI systems can discover vulnerabilities much faster than traditional methods.

The need for automated defense is growing as new AI systems now find vulnerabilities more quickly than traditional processes can keep up. The widely discussed Claude Mythos Preview from Anthropic reportedly found thousands of vulnerabilities and was only released in controlled access due to its capabilities.

Security researcher Himanshu Anand recently emphasized the importance of reducing the 90-day grace period for software vendors, as language models accelerate bug discovery and can quickly turn patches into functional exploits. Google shares this concern, noting that attackers armed with AI can exploit vulnerabilities in just a few hours.

Francis deSouza, COO of Google Cloud, highlighted the difference between Google and its competitors, stating that Google aims to provide ready-to-use patches rather than just simple alert lists. However, questions remain regarding the reliability of self-generated patches and the management of potential errors when the system modifies code live. Google cites Accenture, Deloitte, Netenrich, PwC, and TENEX.AI as partners in implementing this technology.