US considers limiting China's access to AI software

The Biden administration plans to restrict China's access to advanced AI models, including those used in ChatGPT and other AI platforms.

The Commerce Department is considering new regulations to restrict the export of closed-source or proprietary AI models. These models keep their training data and software confidential. Other measures have been taken in the past two years to stop the transfer of advanced AI chips to China, aiming to hinder Beijing's development of this technology for military purposes. Regulators will still find it challenging to keep up with the industry's rapid advancements.

The Department of Commerce chose not to respond. A request for comment was not immediately answered by the Chinese Embassy in Washington. Nothing is now preventing American AI behemoths like Alphabet's Google DeepMind, Microsoft-backed OpenAI, and competitor Anthropic—which have created some of the most potent closed-source AI models—from selling them to practically anybody on the globe without being subject to government regulation.

Public and private researchers are worried that the United States' enemies could use advanced models to carry out cyberattacks or develop strong biological weapons. According to sources, the United States is considering an export ban on AI models. They might use a threshold based on the processing power required to train a model. This threshold was mentioned in an executive order on AI published in October last year. Developers must send their AI model development plans and test results to the Commerce Department when reaching that stage.

Based on that computing power level, two U.S. officials and an additional source briefed on the conversations may decide which AI models would be subject to export restrictions. Since the specifics have not been made public, they declined to be identified. It would probably only apply to unreleased models. According to EpochAI, Google's Gemini Ultra is close to passing the threshold, but no model is believed to have done so yet.

The sources emphasized that the agency is far from completing a regulation proposal. Despite the challenges, the US government is considering enforcing strict regulations on rapidly advancing technology. This shows an effort to address gaps in stopping Beijing's AI ambitions.