The Congress legislation to export control AI models sparked panic among Chinese H1B visa holders in the U.S.

The recently advancing Enforce Act in the U.S. Congress has caused quite a stir among Chinese communities in the U.S. and China.

On May 8, the Act was introduced by Michael McCaul, Chair of the House Foreign Affairs Committee, and co-sponsored by several House representatives from both Reps and Dems. The “Enhancing National Frameworks for Overseas Restriction of Critical Exports Act” (ENFORCE Act) makes three amendments to the 2018 Export Control Reform Act. It authorizes the U.S. Commerce Secretary, after consulting with the Secretary of State, Defense Secretary, and Energy Secretary, to impose export controls on certain “AI systems” and restrict Americans from collaborating with foreign countries in developing these systems. The Act was passed by the committee after a ten-minute markup with a 43:3 vote and reported to the House panel on 22 May.

The Act’s “temporary definition” of “AI systems” includes proprietary or closed-source models and does not exclude open-source models like Llama 3. On the same day, Reuters reported, citing "three people familiar with the matter," that the Commerce is considering new regulations to limit the export of proprietary or closed-source AI models because their software and training data are confidential. The work is still in the drafting rules stage.

This news suggests that Commerce mainly considers export controls to closed-source models. Due to geopolitical and regulatory concerns, major U.S. AI companies have not entered the Chinese market. OpenAI does not offer services to users in China, and its first Asia-Pacific office is in Japan, not Hong Kong. In fact, U.S. companies have effectively “self-controlled”, making export controls on closed-source models somewhat redundant.

Is the Commerce also considering restricting the export of open-source models, which could impact China’s AI ecosystem? Historically, the U.S. has been hands-off with open-source technology, only regulating open-source encryption software with military intelligence significance. Due to strong support from the U.S. tech community and the role of open-source in innovation and market competition, the government has been cautious in regulating open-source models. However, the U.S. government is closely monitoring the use of American open-source models by Chinese companies, who fine-tune these models to build their own large models and offer services. They are evaluating the next steps.

On October 30, 2023, President Biden issued an AI executive order directing the Commerce Secretary to consult with the Secretary of State on the potential risks, benefits, impacts, and appropriate policies and regulations for open-source dual-use foundational models, seeking public input and issuing a report.

Weeks ago, China released a draft national standard on AI safety. The standard allows providers to use open-source models like Llama 3 and fine-tune them to offer services, which was somewhat discouraged by a previous technical document.

However, the Enforce Act has caused a stir in China not because of potential open-source model export control but because of its impact on attracting U.S. AI talent and Chinese H1B visa holders to the U.S.

The Enforce Act adds a restriction at the end of Section 1753(a) of the 2018 Export Control Reform Act, allowing the President to restrict “United States Persons” from participating in activities related to specific restricted AI systems and technologies deemed critical to U.S. national security.

Since the Enforce Act does not define “United States Persons,” it follows the usual legal definition, including U.S. citizens, PRs, anyone within the U.S., and entities registered in the U.S. Similar to the October 7, 2022 semiconductor export controls, under the Enforce Act, “United States Persons” involved in supporting restricted AI systems in China will likely need a BIS license, which may be presumptively denied.

This will affect U.S. citizens or PRs working on controlled AI-related projects at Chinese companies or their partners. Unless these individuals renounce their U.S. citizenship (which is unlikely and complex), they cannot continue their work. Many U.S. citizens or PRs working on controlled AI projects in China may be forced to resign and leave China if the Act is in effect and specific rules are issued by Commerce.

Recently, Microsoft moved some AI research staff from China to the U.S., causing a stir in China. Some Chinese policy observers believe Microsoft might have gotten wind of the Enforce Act, anticipated its passage, and taken preemptive action.

The Enforce Act also affects "Chinese persons" in the U.S. For example, Chinese employees working at U.S. companies, many of whom hold H1B visas and are AI tech talents, might be restricted under "deemed export" rules due to their nationality and ties to China. This means they could be barred from working on large models or even broader AI fields like machine learning and deep learning, forcing them to switch roles or leave their jobs. Future job prospects for Chinese AI talent in the U.S. could become very difficult. Chinese AI companies in the U.S. will also struggle to hire talent, as joining a Chinese company could bring legal risks under U.S. export control laws.

This news has sparked panic among H1B visa holders in the Chinese-American community on WeChat. They discuss the potential impact of the Act on their careers and call on each other to write to lawmakers to express their opposition. Some H1B holders with licensing experience say that U.S. employers might delay hiring due to the license requirements, making them reluctant to hire Chinese students. A former Intel employee has noted that around 2018, it took months for Intel to get approval to hire Chinese H1B holders, effectively creating a job ban.

Chinese policy observers believe that the Enforce Act's restrictions on Americans will negatively impact China's AI talent development in the short term, as many U.S. citizens or PRs working in Chinese AI companies might have to resign or leave. However, they also believe this might not be bad for China in the long run. Studies show that top AI researchers globally have low cross-border mobility and prefer to stay in their home countries. If the U.S. doesn't balance national security with talent attraction, it might force many Chinese AI talents in the U.S. to return to China or deter those considering a move to the U.S. from leaving China.