The U.S. Department of Commerce is considering new regulatory measures to restrict exports to China of proprietary or closed-source AI models whose software and data it learns from are kept secret. This was reported by Reuters, citing its own sources, UNN reports.
Details
Currently, nothing prevents American AI giants such as Microsoft, OpenAI, and Alphabet, which have developed some of the most powerful closed-source AI models, from selling them to almost anyone in the world without supervision.
Government and private researchers are concerned that U.S. adversaries could use these models, which mine vast amounts of text and images to summarize information and create content, to conduct aggressive cyberattacks or even create powerful biological weapons.
Reuters sources said that in order to develop export controls on AI models, the US may refer to the threshold level contained in the AI executive order issued last October, which is based on the amount of computing power required to train the model. Upon reaching this level, the developer is obliged to inform about its plans to develop AI models and provide the results of testing to the Ministry of Commerce.
This computing power threshold could be the basis for determining which AI models would be subject to export restrictions, according to two U.S. officials and another source briefed on the discussion. They declined to be identified because the details are not public.
If it is applied, it is likely to limit exports to only those models that have not yet been released, as none of them have reached this threshold yet, although Google's Gemini Ultra is considered close to it.