Vals study finds open models DeepSeek and GLM nearly match top closed AI at simulated network hacking
🤖 AI-generated content — The title and summary were produced automatically by artificial intelligence, without human editorial review.
Measurement firm Vals ran multi-step intrusion tasks on mock corporate networks and concluded open-weight systems approach the strongest closed frontier models in capability. Researcher Nikil Ravi says downloadable models are catching lab-only systems, weakening the case that restricting releases alone slows dangerous skill spread. Anthropic tied limits such as April’s Claude Mythos Preview to buying time before advanced misuse spreads broadly; its frontier red-team lead Logan Graham told WIRED that Project Glasswing aims to put full new-model capability in trusted defenders’ hands first.
Continue in the app — vote & join in ➔Source: Wired · via ahirlevel.hu