The June Order Directed Officials to Develop a Benchmarking Process Within 60 Days, a Deadline That Has Passed

OpenAI, Anthropic, and Google met with White House officials Tuesday to review the administration's new framework for assessing advanced AI models' cybersecurity risks. The framework, outlined in a June executive order, would require companies to share models with the government up to 30 days before public release, though it's described as voluntary. Key questions remain: how committed the administration is to its light-touch regulatory stance, which models the framework will cover, and what authority the government will have over AI companies.

Jessica Ji, an analyst at Georgetown's Center for Security and Emerging Technology, said officials seem to be creating a soft compliance regime. The framework's scope may exclude open-source models for now, but plans could change. The June order directed officials to develop a benchmarking process within 60 days, a deadline that has passed.