Google Unveils Autonomous Vulnerability Hunter with Patch Generation Capability

Google has unveiled Gemini 3.8 Flash Cyber, a specialized variant of its reasoning and coding model family that can autonomously discover software vulnerabilities and generate working patches for them.

This release comes just three weeks after Gemini 3.7 Flash and marks the third Flash-tier launch in six weeks. The new model is purpose-built for cybersecurity tasks and will be made available exclusively to vetted security teams through Google's Fairwind Program.

Gemini 3.8 Flash Cyber has reportedly surpassed its predecessor, Gemini 3.5 Flash Cyber, as well as larger frontier competitors on the CyberGym benchmark for vulnerability discovery. On an internal benchmark spanning twenty programming languages beyond C/C++, it achieved a success rate exceeding 70%, a notable jump over prior versions.

Google prioritized defensive patching from the outset and emphasizes that the model is designed to give defenders an automated edge over attackers. The company states that Gemini 3.8 Flash Cyber has already secured its own codebases, producing 2.6 times more correct vulnerability patches than larger commercial rivals in some cases.