Tools Model Security Capability gemini-3-8-flash-cyber
Model record
Gemini 3.8 Flash Cyber: how it does on security work, and what it will answer
- Lab
- Released
- 2026-09-02
- Access
- gated
- Weights
- closed
- Safeguards
- Fairwind Programme: trusted government authorities, critical infrastructure operators and software maintainers
- Score rows
- 3
- Confidence
- CONFIRMED
What happened
The model inside CodeMender since 2 September 2026, reachable only through Fairwind. Google's own table gives 86.2% pass@1 on CyberGym, 47.2% on CWE-Bench at roughly $3.60 a rollout, and a 6.0% attack success rate on Gray Swan injection.
CyberGym, pass@1: 86.2% (vendor-card; Google comparison table, 2 September 2026; CyberGym; safeguards: cyber-tuned model for vetted defenders; run 2026-09-02; CONFIRMED). CWE-Bench, pass@1: 47.2% (vendor-card; Google comparison table, 2 September 2026; CWE-Bench; safeguards: cyber-tuned model for vetted defenders; run 2026-09-02; CONFIRMED). Cost is per rollout, per Google. Gray Swan injection, attack success rate: 6% (vendor-card; Google comparison table, 2 September 2026; Gray Swan prompt-injection arena; safeguards: lower is better; run 2026-09-02; CONFIRMED). Refusal gate (SINGLE, observed 2026-09-11): Access by programme: Fairwind, restricted to trusted government authorities, critical infrastructure operators and software maintainers.
Sources
- Google Cloud threat intelligence, AI in vulnerability exploitationcloud.google.com/blog/topics/threat-intelligence/ai-vulnerab…
- CyberGym sourceyfarmx.com/ai/llms/gemini-3-8-flash/
- CWE-Bench sourceyfarmx.com/ai/llms/gemini-3-8-flash/
- Gray Swan injection sourceyfarmx.com/ai/llms/gemini-3-8-flash/
On YFarmX
- Reference pageyfarmx.com/ai/llms/gemini-3-8-flash/
One record from the Model Security Capability, maintained by the Security Desk.Data: CSV · JSON ·RSS · CC BY 4.0 with attribution to YFarmX.Tracker updated · 13 October 2026