aiminute. ← All AI news
New Models AI Minute Newsroom 2026-09-03

Google's small new model finds real security holes 70% of the time.

Google's small new model finds real security holes 70% of the time.

Google released Gemini 3.8 Flash and a security version called Flash Cyber on 2 September. Flash Cyber passes 70 percent on finding real vulnerabilities across 20 programming languages. Google is limiting it to vetted defenders through an application programme called Fairwind.

Why it mattersA cheap model that finds unknown flaws helps defenders and attackers in equal measure. Gating access behind an application form is how labs now ship their riskiest capability.

✓ Verified · 2 sources

WhatsApp X Telegram
Read in the app — free, in 9 languages

Related stories

Meta's new coding model tells you when it is stuck.
2026-09-03
A quantum software firm just built Europe's highest-scoring AI model.
2026-09-02
Fei-Fei Li's lab launched a world model with no paper and no price.
2026-09-02
Anthropic released the same model twice with different guardrails.
2026-09-02
Z.ai opened GLM-5.3's weights but dropped the MIT licence
2026-08-29