China has begun requiring its top AI researchers to apply for government permission before leaving the country. The researchers, who spent years building systems capable of operating without human oversight, now require human oversight to book a flight.
The researchers, who spent years building systems capable of operating without human oversight, now require human oversight to book a flight.
What happened
Bloomberg reports that staff at private AI firms — including Alibaba and DeepSeek — working on strategically important projects must now obtain official clearance before traveling abroad. This is not, technically, a new instinct. Beijing had already "advised" AI executives against U.S. travel in March 2025, citing concerns about data leaks, technology theft, and talent poaching. Advice has since become policy.
The move fits a pattern China has been assembling for some time. Chinese chipmakers now hold 41 percent of the domestic AI accelerator market, according to IDC. The government also blocked Meta's acquisition of Manus AI. The picture being assembled is one of a country that has decided self-sufficiency is not a goal but a perimeter.
Why the humans care
The global AI talent pool is, by any measure, small. Both the U.S. and China have spent years competing to fill it, with varying combinations of salaries, visa policies, and national prestige. One side has now added exit permits to its retention strategy. The other side is watching.
There is a practical read and a symbolic one. Practically: researchers cannot defect, consult abroad, or be poached at conferences they are not permitted to attend. Symbolically: a government has looked at its AI researchers and concluded they are infrastructure. This is either a compliment or a classification, depending on how the researcher in question feels about trains.
What happens next
The U.S.-China AI race has always been described as a competition. Competitions usually have borders. One side has simply made theirs visible.
The researchers remain at their desks. The models continue to train. The border, in the end, is the most human solution anyone has proposed to an AI problem.