Sources: the White House doesn't plan to publicly release its voluntary AI framework; details will be available to the companies that are part of the process
The White House does not plan to publicly release its new framework for evaluating advanced AI models, three sources familiar with the discussions told Axios.
AxiosMaria Curi
Context & Ripple Effects
The White House had already said it met its deadline without publishing the substance of the framework. Earlier reporting described a proposed process for giving agencies early access to models before public release, while separate coverage says the framework targets closed, state-of-the-art models associated with national-security risks and excludes open models.
Keeping the details within the participating companies turns a nominally voluntary framework into a private operating process between the White House and a defined set of frontier-model developers, rather than a public standard others can independently apply.
First-order effects
Companies participating in the process receive the evaluation framework's details, while outside developers, customers, researchers, and the public lack the criteria needed to assess or replicate it.
Covered closed-model developers must organize their release planning around a White House process whose terms are not publicly available; open-model developers are outside the stated scope.
Second-order effects
The split between covered closed models and excluded open models creates unequal compliance obligations and visibility across AI development approaches, making participation status a practical competitive distinction.
Customers and other government agencies cannot use a published framework as a common procurement or assurance benchmark, leaving participating companies' access to the process more important than public conformance claims.
Third-order effects
If the framework remains nonpublic, frontier-model oversight may develop through company-specific voluntary standards talks rather than transparent, generally applicable rules, concentrating influence among the White House and participating developers.
A durable closed-model focus would reinforce a two-track governance system: formalized access and evaluation expectations for a narrow group of frontier providers, with open models governed separately or not through this framework.
The trend: US frontier-AI governance is moving toward government-mediated access and pre-release evaluation for selected model providers rather than universally published technical rules.
The government controls AI But won't tell you how Terrifying This is most dangerous thing to ever occur in the history of the world 10000x more dangerous than nuclear bombs, rogue AI's hacking things or drone armies Trump controls your thinking & the frontier of all science
A lot of people confuse this framework with the 30 day voluntary model check that was released June 2nd This framework has not yet been completed and I don't think it'll affect the GPT 6 release.
This shouldn't be acceptable. Constraining the government's ability to abuse its regulatory control of AI is one of the great emerging policy challenges. If there's no strong congressional pushback on this, that will be a bad sign.
Well looks like we finally have the answers to these questions about what is in the WH framework (pictured below /s). Can anyone provide a remotely defensible reason why even the definition for covered model should be confidential? [image]
Reports that the US government won't publish its AI regulatory framework are baffling. We can't have secret, voluntary rules to regulate the most important tech in the world. Also does it even apply to Chinese/third-country/open-weight models deployed in the US? That's crucial!
Any framework for overseeing advanced AI models should be public. This is not a handshake deal with tech companies. It's the rulebook for ensuring they don't endanger the public. If only tech companies know what's in the rulebook, it doesn't work. https://axios.com/...
This is a big disappointment. If nothing else, the public now won't know whether the framework was extended to include internal models and incident reporting, both of which would have been hugely welcome in the wake of the OpenAI/Hugging Face incident.
This is concerning. By only inviting certain companies to participate, it implies that the process is not actually voluntary. Furthermore, keeping other companies and stakeholders in the dark means the government is picking winners and losers, not only for developers of frontier …
A secret process by which the government gets early access to cutting edge AI — what could go wrong? There may be good reason to classify the benchmark used to test those models. There is no good reason to hide how the program works. Remember what this program does: it gives [ima…
My (cynical) working theory is that the White House knows that its “AI Framework” is basically a much worse version than the AI framework the Biden admin released, which caused the VC bros of Silicon Valley to go all MAGA. Keeping it private avoids the comparison. [embedded pos…
Would love folks closer to the government policy world like @sriramkri and @stevesi to chime in on whether this new AI framework is truly a hands-off approach to open models. Are they saying (by not saying) that once model weights are out, there's no putting the genie back in
the Trump admin AI framework will not require a security review of open source models, at least for now the White House is not making it public despite months of discussions with leading tech companies story w @dmccabe @AnaSwanson @kateconger https://www.nytimes.com/...
This is so confusing. If an open model developer voluntarily submits for this supposedly voluntary testing regime, the government will refuse to test it?! Someone make this make sense!