AnalysisCybersecurityNovember 6, 2026

Open-weight Chinese models near frontier hacking skills, Vals finds

Read original source →wired.com

Vals tested models on multi-step intrusions into mock corporate networks and found DeepSeek and GLM not far behind OpenAI's most advanced cyber model, which isn't public. Researcher Nikil Ravi says open models anyone can download now match gated frontier ones, undercutting lab access limits like Anthropic's Project Glasswing.

People · Nikil Ravi, Logan Graham

1 source

More stories today

Open the live feed