For decades, the people who needed permission to leave China were a fairly narrow group: nuclear scientists, senior state-enterprise executives, prominent university researchers, and the Communist Party officials whose passports were already sitting in a safe somewhere. Bloomberg reported this week, in a story that has now ricocheted through Tom's Hardware, TechCrunch, The Straits Times, RT, India Today and the smaller crypto and tech trade press, that the list has been quietly extended. AI researchers, executives and start-up founders at private companies including Alibaba and DeepSeek have been told they will now need government approval before traveling abroad.
The reporting agrees on the shape of the policy and disagrees, interestingly, on its temperature. The Straits Times and India Today frame it as a step-up of an existing, advisory regime: in March 2025, The Wall Street Journal reported that top AI founders were merely being advised against U.S. travel, while some private-sector engineers were already required to file flight plans with the authorities. Briefs.co goes further, claiming that DeepSeek began collecting researchers' passports in March, in the immediate aftermath of its R1 model surprising the world by matching American benchmarks on a fraction of the compute. RT, predictably, presents the move as part of a coordinated drive for self-reliance. Tom's Hardware reads it as the act of an authoritarian system that fundamentally does not understand its own talent.
Beneath those framings sits a fact that all the outlets converge on: Beijing is now treating high-end AI researchers the way other countries treat fissile material. The trigger, per multiple reports, was the Manus episode. Manus, a Chinese AI start-up that had relocated its corporate home to Singapore, became the subject of a $2 billion acquisition bid from Meta. Chinese regulators blocked the deal, the Financial Times reported the co-founders were barred from leaving the country during the probe, and the National Development and Reform Commission then quietly instructed leading AI firms, including Moonshot AI, StepFun and ByteDance, to refuse American capital without prior approval. Travel control of the people is the logical extension of capital control of the firm.
What gives the story its weight is timing. Stanford's index now puts the performance gap between the top American and Chinese frontier models at 2.7 per cent, down from 31 per cent in 2023. China files about 70 per cent of global AI patents and installs industrial robots at nine times the United States' rate. Beijing has decided that, having closed the gap, the cheapest insurance policy is to keep the people who closed it physically inside the country. The implicit theory of the policy is that loyalty is geographic.
The risk is that the theory is wrong. The Tom's Hardware comment thread, which is rarely the place to look for sober geopolitical analysis but in this case is unusually clarifying, runs heavily to one observation: if you tell ambitious AI engineers their passports may be the next thing taken, the ambitious ones leave now. Briefs.co notes that the flow of AI talent from China to the United States has already fallen 89 per cent since 2017, so there is less to lose than there once was. But there is also less margin for error. China's AI lead is concentrated in a small number of laboratories and a smaller number of people. Each one who reads the news this week and decides to relocate to Singapore, Dubai or Munich while they still can is, in effect, a small piece of strategic policy walking out the departure gate.
There is a deeper, less comfortable echo. American security officials spent 2025 detaining at least three Chinese researchers on espionage charges, and the United States' own export controls now treat people, in the form of "deemed exports" of knowledge, as license-controlled goods. The two regimes are converging on the same instinct from opposite directions: that in an industry where the talent is the technology, the border is the natural place to fight the war. Whether that instinct produces a more stable rivalry or a more brittle one is the question that does not yet have an answer. What is clear is that the question is now operational, not theoretical. Somewhere in Hangzhou this week, a senior researcher at Alibaba had to ask permission to attend an academic conference. Whatever the answer was, it changed the shape of the AI race.