Open-weight AI models are catching up to the frontier. The safety gap remains. 

New research shows that powerful open-source AI models are becoming as capable as frontier commercial models, but lack important safety protections. This raises concerns that safety standards may fall behind as AI becomes more widely available and accessible.