At a glance: Google’s new specialized model Gemini 3.5 Flash Cyber identifies security vulnerabilities in code more accurately than competing models and is currently available only to governments and select partners.
Google DeepMind has developed Gemini 3.5 Flash Cyber, a specialized AI model for identifying and fixing security vulnerabilities in source code. The system is being provided exclusively to governments and select partners through the CodeMender platform as part of a controlled pilot program.
Gemini 3.5 Flash Cyber is based on Google’s Gemini 3.5 Flash and is designed as a lightweight model. It is intended to perform source code analysis at high frequency and with low execution costs. The model not only identifies vulnerabilities but also verifies them and offers remediation measures.
In internal evaluations, the model demonstrated superior results in detecting unknown vulnerabilities. In tests of the V8 JavaScript engine, Gemini 3.5 Flash Cyber identified 55 confirmed security issues, while Gemini 3.5 Flash found 47 and Anthropic’s Claude Opus 4.6 discovered 36. Ten of the identified vulnerabilities were found exclusively by the new specialized model. Additional tests were conducted on more complex codebases such as Google Chrome and Apple Safari.
Google deliberately restricts access to the model to minimize abuse risks. Security leads Raluca Ada Popa and Four Flynn justify this by noting that automated vulnerability discovery represents a double-edged risk. The model runs exclusively within the CodeMender environment, where specific safeguards are in place. This enables defensive analysis while preventing malicious activities.
The planned expansion of the pilot program is intended to proceed gradually. This will give security professionals a head start in finding and fixing critical vulnerabilities before they can be exploited.
Source: www.it-daily.net · Published 22 July 2026
Lumi AI News — AI-assisted curation in accordance with Art. 50 EU AI Act. Paraphrase and classification by Lumi News Pipeline v1.7.3.