DeepMind introduced Gemini 3.5 Flash Cyber, a cybersecurity model designed to identify, validate, and patch vulnerabilities more efficiently than previous versions. The model is built on top of the 3.5 Flash foundation and is fine-tuned for cybersecurity tasks, offering a cost-effective and scalable solution for securing codebases. According to DeepMind, this model enables faster vulnerability detection and response, critical for addressing growing cyber threats. The release follows years of investment in automated vulnerability discovery through tools like CodeMender, which has already been used to secure major codebases.
Gemini 3.5 Flash Cyber is particularly suited for scanning large codebases and analyzing numerous code paths, which is essential for finding deep-seated flaws. By invoking the model multiple times, CodeMender can generate comprehensive reports on vulnerabilities without relying on a single, expensive call to a large language model. This approach allows for frequent and scalable scans, making it ideal for time-sensitive processes and continuous integration pipelines. The model's efficiency and affordability make it a strong alternative to larger, more costly cybersecurity models.
DeepMind emphasized that the dual-use nature of the technology led to a limited-access pilot program, with 3.5 Flash Cyber initially available to governments and trusted partners through CodeMender. This strategy aims to give frontline defenders an early advantage in identifying and mitigating critical vulnerabilities before they can be exploited. The company also plans to expand access over time, ensuring broader adoption of the model.
Source: deepmind
