US government introduces voluntary framework for testing risky closed-source AI models, excludes open models
According to Axios, the Trump administration has introduced a voluntary framework for testing the security risks of AI models. It applies only to closed-source models with cutting-edge capabilities that pose a national security risk; open models are excluded, and key terms are undefined.
According to Axios, the administration of US President Trump has finalized a framework for voluntary testing of the security risks posed by advanced AI models. The framework was developed under an executive order that the president signed in June, which asked AI companies to share their frontier models with the federal government before release due to cybersecurity concerns. Anthropic, OpenAI and Google reportedly attended a White House briefing on the final version of the framework, but the administration does not plan to publish the framework itself.
The framework applies only to closed-source models that have the most advanced capabilities achieved (“state-of-the-art”) and pose a national security risk. For these models, it sets a 30-day period during which the government has the opportunity to review a new model before its release. Participation is voluntary, and companies are not required to comply with the framework.
The framework explicitly excludes open models — meaning those that anyone can download and examine their internal components — and states that it cannot be used to restrict open models after their release. At the same time, it does not define exactly what “state-of-the-art” or “national security risk” means, which, according to the source, conflicts with the stated purpose of the framework to assess these risks.
According to the source, leading AI labs such as OpenAI and Anthropic are actively seeking guidance on how to release models without triggering government restrictions. The lack of clear definitions for key terms means continued uncertainty, which could be a problem particularly for smaller AI providers seeking a good relationship with the administration.
Why it matters
The framework introduces a voluntary option for government review (30 days) before the release of closed-source frontier models that pose a national security risk, but without legal enforceability. Because it does not define what “state-of-the-art” or “national security risk” means, there is uncertainty over who the rules actually apply to — which, according to the source, places the greatest burden on smaller AI companies trying to meet government expectations, while large labs actively seek ways to avoid restrictions. Open models are entirely exempt from the framework and cannot be restricted under it after release.
Relevant practical impact
What this means
For a business
Companies developing closed-source AI models with the most advanced capabilities can voluntarily undergo a 30-day government review before release due to national security risk; because the framework does not define key terms, uncertainty could particularly affect smaller AI providers trying to comply with government expectations.
Risks and complianceCheck the original
Event sources
only one source so far · 1 publisher, 1 independent. We count feeds from the same owner only once.