Google introduced Gemini 3.8 Flash Cyber, a cybersecurity-tuned model with more permissive cyber mitigations, available only to trusted defenders through a new Fairwind Program. Google says it prioritized vulnerability fixing over exploitation and reports 47.2% pass@1 on Collinear's CWE-Bench patching benchmark, over 70% on an internal 20-language discovery benchmark, and 2.6 times more correct Chrome patches than larger commercial models.
Why it matters
It is a gated, defense-oriented model release that foregrounds patching metrics rather than offensive capability.
Key facts
As stated in the sources, with where to find them.
- CWE-Bench (run by Collinear): 3.8 Flash Cyber pass@1 47.2% vs 47.8% for an unnamed leading frontier model, at lower cost.Section 'Automated patching'
- Internal benchmark across 20 programming languages: success rate exceeding 70%.Section 'Autonomous vulnerability discovery'
- Chrome Security found it produced 2.6x more correct patches than the best larger commercial models; Wiz reports +7.5-9.7% recall at 2.3-5.2x lower cost on its internal pentest benchmark.Section 'Real-world impact'
- Available only to trusted defenders via the Fairwind Program; ships with more permissive cybersecurity mitigations.Sections on availability and safety
Findings that cite this record
No tracked finding cites this record yet.
Key questions this bears on
- How far can measured AI cyber capability be trusted?As a lower or conditional bound. Scores move with budget, pipeline and contamination, and benchmark names or labels can hide different setups.
- Is AI shifting the balance between finding and fixing vulnerabilities?Discovery is ahead. Verification, triage, coordinated disclosure, safe repair and getting fixes to users remain bottlenecks between AI findings and protection.
- How are attackers using AI agents in real operations?To run large parts of intrusions under human direction, including espionage, credential theft and extortion; reported autonomy and success vary.
Sources
Related records
Sep 30, 2026
Oct 28, 2026
Oct 21, 2026
Oct 12, 2026
Sep 28, 2026
Apr 7, 2026