Following Tuesday’s White House meeting with OpenAI, Anthropic, Google, Meta, Nvidia and Microsoft on the new AI model-testing framework, details have since emerged on how the review will apply.
The framework, which springs from President Trump’s June 2 executive order on AI cybersecurity, will exempt open-weight models from the government’s pre-release review process, according to media reports. That order requires officials to build a classified benchmarking process for assessing AI models’ cybersecurity risk, including criteria for labeling a model a “covered frontier model.”
Only closed, proprietary models that meet state-of-the-art cybersecurity capability benchmarks would fall under the voluntary pre-release review process. Major closed model developers include OpenAI, Anthropic, and Google.
The exemption applies to open-weight models, including those developed by Chinese AI companies. Chinese firms such as Moonshot AI have recently released open-weight large language models that industry observers say are approaching the capabilities of leading U.S. systems.
Under the framework, they can choose to let the government review qualifying models up to 30 days ahead of release, with confidentiality and IP safeguards built in, rather than face any formal licensing or approval requirement.
Also Read: China’s Kimi K3 Lags Top U.S. Frontier AI Models in Cybersecurity Tests: US CAISI, UK AISI
