According to Axios, the U.S. government announced on Monday local time that it had completed a voluntary assessment framework for advanced AI models on schedule. However, the White House has not revealed the specific content of the framework, nor has it specified which agencies or companies have reviewed the document, and it has also not provided a timeline for when companies will begin adopting it.

This framework originated from a presidential executive order signed on June 2, aiming to provide AI developers with a structured process to help them collaborate with the government in determining whether models under development fall within regulatory scope. The framework should clarify the confidentiality requirements, cybersecurity requirements, internal risk management requirements, intellectual property protection, and non-disclosure provisions that the government must follow when obtaining relevant models up to 30 days before their release. It should also define which "trusted partners" can gain early access to the model.

What should be public remains hidden, and what should be confidential is even more so.

Notably, the executive order explicitly classified the benchmark testing process for evaluating an AI model's advanced cyberattack capabilities as confidential, and the threshold parameters for applicable models are also considered confidential information, disclosed only to AI developers and researchers "under appropriate circumstances." However, the voluntary framework itself was not classified as confidential, so policymakers and observers had previously expected to see the details—yet the White House has still not made them public.

A White House official's response was rather telling: "Even if certain information is not classified, it does not mean we will disseminate it to everyone." He also stated that the government is engaging with "more than just Anthropic, OpenAI, and Google" in terms of industry partners. By the deadline, these three AI labs had already provided feedback on the framework draft to the government.

AI companies are eager to obtain clear guidance as soon as possible to determine whether their models under development may be subject to the framework. According to informed sources, the White House plans to hold staff-level meetings with relevant companies on Tuesday local time to review the framework. When the world's most powerful AI models will face what kind of regulatory rules, even the companies that provided feedback remain in the dark, and this "completed but not visible" framework itself may already reveal a lot.