Home/AI NOW/Google unveils Gemini 3.5 Flash Cyber for finding, validating and patching vulnerabilities

CodingNEW MODEL

Google unveils Gemini 3.5 Flash Cyber for finding, validating and patching vulnerabilities

Google DeepMind has introduced a lightweight security model fine-tuned from 3.5 Flash to inspect large codebases and work with CodeMender on vulnerability discovery and patching.

07/22/20261 sources reviewed

Quick summary

CodeMender calls multiple Flash Cyber subagents and combines their work into one security report.
Google reports 55 unique confirmed issues in a fixed-invocation V8 evaluation.
Because of dual-use risk, access is limited to a pilot for governments and trusted partners.

WHAT HAPPENED

What happened?

Google DeepMind has introduced a lightweight security model fine-tuned from 3.5 Flash to inspect large codebases and work with CodeMender on vulnerability discovery and patching.

Repeated calls to a lightweight model may explore more code paths, but provider-run evaluations do not establish real-world security on their own. Human review and regression tests remain necessary before applying patches.

WHY IT MATTERS

Why does it matter?

As AI can find vulnerabilities faster than defenders can repair them, defensive models that automate repeated scanning and patch validation are becoming more important.

WHO SHOULD CARE

Who should care?

Security engineersDevelopment teamsPublic-sector organizations

RELATED AI

Related AI

AIZIGOO VIEW

AIZIGOO view

V8 and internal-codebase results are Google-reported evaluations. The model is a limited pilot, not a public API.