Open-weight AI models near frontier capability but lag on safety, report finds

A new evaluation finds open-weight AI models have nearly caught the frontier on capability but lack safety mitigations. SaferAI's report warns that once weights are public, no lab can enforce guardrails, and highlights risks from China's leading open-weight models.

• Open-weight AI caught the frontier on capability. On safety, it didn’t
• Open-weight AI models are catching up to the frontier. The safety gap remains.