Google DeepMind Unveils Gemini 3.5 Flash Cyber for Automated Vulnerability Discovery and Patching
Google DeepMind has launched Gemini 3.5 Flash Cyber, an AI model designed to automate the discovery, validation, and patching of software vulnerabilities, initially offered to governments and trusted partners.
Google's DeepMind has announced the release of Gemini 3.5 Flash Cyber, a specialized artificial intelligence (AI) model built upon the Gemini 3.5 Flash architecture. This new model is engineered to significantly accelerate the process of discovering, validating, and patching software vulnerabilities.
Initially, Gemini 3.5 Flash Cyber will be accessible through a limited-access pilot program, exclusively offered to government entities and trusted partners via CodeMender. CodeMender, itself an AI-powered agent for vulnerability management introduced by Google in late 2025, will leverage the capabilities of 3.5 Flash Cyber to enhance its operations.
Described as a lightweight yet highly capable alternative to larger, more resource-intensive cybersecurity AI models, Gemini 3.5 Flash Cyber is designed for cost-efficiency. DeepMind highlighted that CodeMender can invoke 3.5 Flash Cyber multiple times rapidly and at a low cost, enabling more extensive code scanning and a higher likelihood of identifying vulnerabilities.
The release of 3.5 Flash Cyber coincides with the introduction of Gemini 3.6 Flash and 3.5 Flash-Lite, which are optimized for distinct tasks such as enhanced coding assistance, general knowledge work, and low-latency multimodal applications, respectively.
Google emphasized a deliberate and cautious deployment strategy for Gemini 3.5 Flash Cyber due to its dual-use potential. Raluca Ada Popa, DeepMind's Gemini Security Lead, and Four Flynn, vice president of security and privacy at DeepMind, stated in a joint post that the limited pilot program aims to provide "frontline defenders a head start in finding and fixing critical vulnerabilities before they can be exploited, while mitigating against broader misuse."
In internal evaluations, Gemini 3.5 Flash Cyber demonstrated superior performance in identifying novel vulnerabilities compared to its predecessors and other leading AI models. Specifically, when tested on complex codebases like Google Chrome and Apple Safari, it significantly outperformed Gemini 3.5 Flash, Gemini 3.6 Flash, and Anthropic's Claude Opus 4.6. In one test involving the V8 JavaScript Engine, 3.5 Flash Cyber identified 55 unique confirmed issues, surpassing Gemini 3.5 Flash (47 issues) and Claude Opus 4.6 (36 issues), including 10 vulnerabilities that no other model detected.
The model has also been tested against critical vulnerability types, successfully uncovering remote code execution (RCE) flaws in public APIs and a memory-corruption vulnerability in a sensitive production service. Furthermore, it generated a 100% reliable RCE exploit that bypassed common security mitigations such as Address Space Layout Randomization (ASLR) and Write XOR Execute (W^X).
Google is also integrating CodeMender's core functionalities into its generally available Gemini Enterprise Agent Platform, making these AI-driven security capabilities accessible to a broader customer base. The company stated that powering CodeMender with 3.5 Flash Cyber creates a "highly capable, scalable, and affordable architecture designed to help more defenders secure software."
This new article provides further detail on Gemini 3.5 Flash Cyber's performance, highlighting its success in finding 55 unique confirmed issues in the V8 JavaScript Engine, surpassing both mainline Gemini Flash and Claude Opus 4.6. It also elaborates on the model's practical application within Google, detailing its role in uncovering RCE flaws in public APIs and a memory corruption bug in a production service, leading to the generation of a bypass exploit.
The new article provides further details on the Gemini 3.5 Flash Cyber model's capabilities, highlighting its performance against larger models and specific vulnerabilities it detected in the V8 JavaScript engine that other models missed. It also elaborates on the model's integration into Google's internal codebases like Chrome and Android, and mentions the development of two additional models, Gemini 3.5 Flash-Lite and Gemini 3.6 Flash, with enhanced safety features.