The Trump administration has unveiled a framework for evaluating cybersecurity risks associated with advanced artificial intelligence systems, but the plan conspicuously avoids assessing openly available models. According to reporting, the voluntary guidelines specifically exclude open-source AI systems that anyone can freely download and examine, and the framework contains explicit language preventing it from being applied to restrict such models after their release.
The testing framework emerged following an executive order signed by President Trump in June that called on AI developers to submit their most advanced models to federal review before making them public. Major AI companies including Anthropic, OpenAI, and Google participated in a White House briefing on the finalized framework, though administration officials have not disclosed the specific details to the public. The plan grants the government a 30-day window to assess newly developed closed-source models that demonstrate cutting-edge capabilities and potential national security concerns.
Despite these parameters, significant ambiguities undermine the framework’s effectiveness. Notably, the guidelines fail to provide concrete definitions for critical terms such as “state-of-the-art” capabilities or what precisely constitutes a “national security risk.” This definitional gap creates considerable uncertainty for smaller AI companies seeking to operate in compliance with governmental expectations while remaining voluntary in nature.
Originally reported by Theverge →
About the Author
