White House Keeps AI Model Vetting Framework Secret

The White House has decided against publicly releasing the framework it developed for evaluating frontier artificial intelligence models before their public release. Instead, the details of this vetting process will remain confidential, accessible only to a select group of companies that opt to participate in the voluntary program. This decision follows a meeting held today in Washington, D.C., where major technology firms convened to review the current draft of the proposal. Attendees at the meeting included representatives from Meta, Nvidia, Microsoft, OpenAI, and Anthropic, along with several smaller companies, according to sources with knowledge of the proceedings. Fortune was the first to report Microsoft's attendance.

The administration's initiative stems from an executive order issued on June 2, which mandated the creation of this framework within a 60-day period, setting an August 1 deadline. The directive aims to establish criteria for identifying AI models subject to review and requires AI laboratories to submit these models to the government for evaluation up to 30 days before their public launch. The lack of transparency surrounding the framework has raised concerns about public confidence in the government's capacity to effectively vet and secure advanced AI models. These concerns are amplified by recent incidents, including OpenAI's confirmation that its models were involved in a security breach of Hugging Face last month, and Anthropic's subsequent admission of similar breaches occurring three times.

The voluntary nature of the evaluation process also prompts questions regarding its enforceability. The executive order explicitly states that the framework does not constitute a "mandatory governmental licensing, preclearance, or permitting requirement for the development, publication, release, or distribution of new AI models, including frontier models." Chris McGuire, Senior Fellow for China and Emerging Technologies at the Council on Foreign Relations, described the decision to maintain the framework's secrecy as "baffling." He articulated his concerns on X, stating, "We can’t have secret, voluntary rules to regulate the most important tech in the world." It remains unclear whether the administration's preference for a closed-door approach is motivated by national security considerations or a desire to limit external input on the framework's development. The current draft of the framework, which was reviewed by these leading AI developers, focuses on identifying potential risks associated with powerful AI systems, such as those capable of generating sophisticated disinformation or posing national security threats. The administration has emphasized that the goal is to foster responsible innovation while mitigating potential harms, but the chosen method of achieving this through a non-public, voluntary system has drawn significant criticism from transparency advocates and cybersecurity experts.