OpenAI and Anthropic have reached an agreement to allow the US government to access new AI models before and after they are released to improve security. The AI Security Institute announced Thursday that the move will help assess security risks and mitigate potential problems. It is reported that California recently passed the Frontier Artificial Intelligence Model Safety and Security Innovation Act (SB 1047), which requires AI companies to implement security measures, sparking industry opposition and concerns about affecting small open-source developers.