Google Introduces Gemini 3.5 Flash Cyber for Cybersecurity
Lightweight AI model rapidly detects and patches software vulnerabilities.
1Next-Gen Security Model
Google unveiled Gemini 3.5 Flash Cyber, an AI model that automatically detects and fixes software vulnerabilities. This lightweight solution offers a cost-effective alternative to bulky cybersecurity models, particularly effective when scanning large codebases.
2Limited Access Pilot Program
Due to dual-use risks, 3.5 Flash Cyber is initially available only to governments and trusted partners via CodeMender. The model gives defenders a head start in patching vulnerabilities before exploitation. Access will expand gradually over time.
3Performance Edge
In CyberGym benchmarks, 3.5 Flash Cyber competed with larger models while outperforming them in complex codebases like Chrome and Safari. Engineering teams confirmed it detected remote code execution flaws within 2 hours.
4Unique Vulnerability Discovery
In fixed-invocation tests, 3.5 Flash Cyber identified 55 confirmed issues in V8 JavaScript Engine - surpassing both its mainline version and Claude Opus 4.6. The model excels at finding novel vulnerabilities rather than repetitive patterns.
5Real-World Implementation at Google
The model already patches vulnerabilities in Chrome, Android and YouTube codebases. Google's Cloud Security team leveraged its speed to detect memory corruption in sensitive services and generate reliable exploit code bypassing standard mitigations.
6Cybersecurity Game Changer
With 10 years of OSS-Fuzz data and the OSV.dev database, this technology promises scalable defense. 3.5 Flash Cyber could set new standards in AI-powered security, giving defenders critical advantage against evolving threats.
Original article: DeepMind — Google, Siber Güvenlik için Gemini 3.5 Flash Cyber'ı Tanıttı