The administration confirmed on Monday that it had met the June 2 deadline to finish a voluntary framework intended to evaluate frontier AI models. The framework, created under President Biden's executive order on artificial‑intelligence safety, provides a structure for deciding whether a model under development falls within the order’s 30‑day pre‑release review window.

White House officials said the document itself is not classified, but the benchmarks that gauge a model’s cyber‑capability and the threshold that determines coverage are. Those details are being shared only with developers "as appropriate," the administration added. "Just because things are unclassified doesn’t mean we are going to broadcast them to everyone," an official told reporters.

Industry feedback came from three of the nation’s biggest AI labs—OpenAI, Anthropic and Google—who reviewed a draft of the framework. The administration said it is also consulting with many more companies, though names were not disclosed. A staff‑level meeting with representatives from the tech sector is scheduled for Tuesday to walk participants through the final version.

The executive order, signed in June, mandates a voluntary, 30‑day government preview for any model deemed a "frontier" system. By establishing a set of benchmarks and thresholds, the new framework aims to give the government a clear metric for applying that review. Critics, however, argue that keeping the benchmarks and thresholds hidden undermines the transparency promised by the order. Without public insight into the criteria, companies and observers cannot gauge whether a model will be subject to review until after the fact.

At the same time, the White House rolled out an initiative called Gold Eagle earlier this month. Gold Eagle is designed to coordinate AI‑driven cyber‑defense efforts across federal agencies. The newly completed framework serves as a companion piece: Gold Eagle identifies vulnerabilities in AI systems, while the framework decides which models are powerful enough to require a government review before release.

Policy analysts and AI‑safety advocates had expected the administration to release the framework’s details alongside the deadline, but the White House chose to withhold them. The decision has sparked a debate about whether a framework built on classified benchmarks can satisfy the executive order’s call for openness. Some lawmakers have asked for a congressional briefing, while industry groups are pressing for clearer guidance on how the thresholds will be applied.

As the staff meeting approaches, developers are preparing to discuss how the undisclosed criteria might affect their product roadmaps. The administration emphasized that discussions with industry about next steps are already underway, suggesting that more information could emerge in the coming weeks.

This article was written with the assistance of AI.
News Factory APP - agentic news to boost your SEO & AEO.