On May 10th, according to national media reports, the US is planning to introduce new AI control measures aimed at restricting the export of proprietary or closed-source AI large models to countries such as Russia, China, North Korea, and Iran.
The plan will specifically target core software of AI models like ChatGPT, setting up export controls for them.
This move is seen as a continuation of a series of control measures taken by the US over the past two years, including restrictions on exporting high-end AI chips to China to slow down its advancement in cutting-edge technology.
Currently, the US government has taken steps to block the flow of AI chips and their manufacturing tools to China, and requires AI companies to report to the government when foreign clients use their services to train AI models that could be used for cyberattacks.
However, there are no clear management measures in place yet regarding the export of AI models themselves.
Major US AI companies like OpenAI, Google DeepMind, and Anthropic have developed advanced AI large models. Without government regulation, these models can be sold to individuals or entities worldwide.
The US government is concerned that these AI models could be used by competitors to launch cyberattacks against the US or develop biological weapons.
Reports suggest that the US may establish export control measures based on the computational power threshold outlined in an AI executive order issued last October. Models meeting this threshold would need to report their AI model development plans to the Department of Commerce and provide test results. However, according to data from Epoch AI, no AI models currently meet this threshold.
Nevertheless, export control thresholds are not set in stone. Sources indicate that the US may eventually establish more detailed requirements, such as data types or potential uses of AI models.
