China Imposes Travel Restrictions on Top AI Researchers at DeepSeek and Alibaba

China has formally extended travel restrictions to senior AI professionals at major private sector companies including DeepSeek and Alibaba, Bloomberg reported on May 26. The policy — which requires individuals deemed strategically important to obtain pre-approval from government authorities before traveling abroad — represents the most aggressive step Beijing has yet taken to treat elite AI talent as a national security asset on par with military technology.

The restrictions apply to startup founders, researchers, and executives working on advanced AI, and mark a significant escalation from informal guidance that had been quietly in place since late 2025. According to people familiar with the policy, the shift from advisory to mandatory is material: what was previously understood as a strong preference has hardened into a formal pre-approval requirement with real enforcement mechanisms.

From DeepSeek to the Whole Sector

The origins of this policy trace to December 2025, when restrictions were first quietly applied to certain executives at DeepSeek’s parent company, High-Flyer. Early reporting described High-Flyer as holding some employees’ passports rather than issuing a formal government travel ban — a practice that sat in a legal grey zone between employer policy and state control.

The expansion announced last week moves beyond individual firms and beyond the informal phase. Beijing has instructed government agencies to begin vetting overseas travel requests from individuals involved in advanced AI work across the private sector. Alibaba — whose AI research division DAMO Academy and large language model portfolio have grown significantly in 2026 — is specifically named in the Bloomberg report alongside DeepSeek.

Neither Alibaba nor DeepSeek has publicly commented on the restrictions.

The Strategic Logic

Beijing’s rationale operates on two axes: preventing talent outflows to U.S. and European AI companies, and preventing Chinese AI researchers from forming relationships or sharing knowledge that could accelerate Western AI development.

On the talent side, the U.S. Department of State issued new guidance in March 2026 to make it easier to issue O-1 visas (for individuals with extraordinary ability) to Chinese AI researchers who can credibly establish that they face pressure not to conduct independent research at home. Several high-profile Chinese AI researchers who had previously worked at ByteDance, Tencent, and Baidu have joined U.S. labs in 2025 and early 2026 — a trend that Beijing appears determined to slow.

On the knowledge transfer side, the concern is more diffuse but arguably more serious. International AI conferences, joint research collaborations, university partnerships, and advisory board memberships are all channels through which technically detailed information moves across national boundaries. A Chinese researcher who spends two weeks at NeurIPS, engages in whiteboard sessions with U.S. colleagues, and returns home has transferred knowledge in both directions — and the direction that flows out is often the direction China would prefer to retain.

Parity Claims and the Containment Pressure

The timing of the restrictions is not coincidental. By multiple external benchmarks, China’s frontier AI capabilities have reached approximate parity with leading U.S. models in several domains. DeepSeek’s R2 series, released in early 2026, outperformed several Western models on mathematical reasoning and code generation benchmarks while running at a fraction of the inference cost — a demonstration that rattled Silicon Valley and triggered months of congressional hearings about whether U.S. export controls were working.

From Beijing’s perspective, the moment of approximate parity is precisely the wrong time to allow unrestricted outflows of the talent that created it. The logic is analogous to the controls the United States has placed on advanced semiconductor manufacturing knowledge: it matters most when you are ahead, because you are protecting something valuable, and it also matters when you are at parity, because you are trying to preserve a lead rather than catch up.

The travel restrictions are the human capital analog to chip export controls — an attempt to prevent the capabilities embedded in researchers’ minds from crossing borders the same way the capabilities embedded in silicon chips cannot.

Compliance, Costs, and Credibility

The practical implementation raises serious questions about compliance and costs to China’s own AI sector.

International collaboration has been a significant accelerant for Chinese AI research. A non-trivial share of the senior talent at China’s top AI labs received their doctoral or postdoctoral training at U.S. or European institutions, maintains co-authorship relationships with overseas researchers, and depends on access to international conferences to stay current with the field. Restricting their travel does not just prevent outflows — it reduces their access to inflows of knowledge that benefit China’s AI development.

There is also the talent retention dimension. For researchers who chose to stay in China despite lucrative opportunities abroad — partly out of patriotism and partly because of the career opportunities China’s AI boom created — the imposition of restrictions on their freedom of movement represents a shift in the implicit deal. Several AI researchers at Chinese firms have expressed concern privately, according to Bloomberg’s sources, about the long-term implications for their career autonomy.

The policy also raises practical questions about international corporate collaboration. Companies like Huawei, Baidu, and Alibaba Cloud have significant international customer bases and need technical staff to travel for client engagements, standards body participation, and regulatory meetings. How the pre-approval process navigates those business necessities without creating operational paralysis remains unclear.

International Reactions and Reciprocity Risks

The restrictions land in a context of escalating mutual suspicion around AI and technology more broadly. The United States has progressively tightened controls on chip exports to China, restricted Chinese AI investment in U.S. companies, and placed several Chinese AI and semiconductor firms on entity lists. The European Union has launched inquiries into Chinese AI platforms’ data practices.

China’s travel restrictions are a different kind of control — one directed inward rather than outward — but they sit within the same logic of treating AI as a national security domain where normal economic openness rules do not apply.

The near-term risk is reciprocity. The United States and its allies could use China’s restrictions as justification for tightening their own controls on Chinese researchers’ access to U.S. academic institutions, conferences, and corporate research programs. That cycle — which researchers on both sides have warned about for years — would slow global AI progress in ways that ultimately hurt both countries, even if the harms were distributed unevenly.

What This Signals

Three things seem clear from this development, regardless of how the enforcement evolves.

First, China has concluded that its frontier AI talent is among its most strategically valuable national assets, and it is prepared to restrict civil liberties to protect them. This is a different order of intervention than controlling chip imports or restricting foreign software.

Second, the pace of AI development has become fast enough that individual researchers’ movements now matter to national security calculations — a situation without precedent in any prior technology wave, including nuclear, aerospace, or semiconductor development, where the relevant knowledge was more tightly held within institutional and governmental structures.

Third, the era of treating AI as a globally open, collaborative scientific enterprise — the culture that produced the transformer architecture, open-weight models, and the shared benchmark infrastructure that the whole field depends on — is ending. The fracture lines are hardening faster than most participants in that collaborative tradition expected or wanted.

For the global AI research community, the most optimistic reading is that restrictions of this kind tend to be imperfectly enforced and gradually eroded by economic and scientific pressure over time. The pessimistic reading is that 2026 is the year when the bifurcation of the global AI ecosystem became effectively irreversible.