Open-weight AI models are catching up to the frontier, analysis finds

Per Semafor, the UK's AI Safety Institute found that the most capable open-weight AI models are four to seven months behind frontier closed models in cyber capabilities. Open-weight models offer benefits—they can be modified for different needs and enable easier collaboration—but safety guardrails can be removed, making powerful cyber and biological capabilities accessible to bad actors. Semafor notes that Boko Haram militants in Nigeria are using mostly closed AI systems to plan attacks and gain tactical advantages. Anthropic's CEO has called for safety testing of all powerful AI models, regardless of whether they're open or closed.

Listen to this story

Hear this and more stories in a personalized audio briefing.

Open The Chonkerton