China extends exit bans to its AI talent in the private sector

China has started preventing its leading AI researchers from leaving the country. According to Bloomberg, authorities have over recent weeks extended travel restrictions to key staff at private companies such as Alibaba and DeepSeek – a group that until now largely escaped the kind of control the state has long used against senior executives at state-owned enterprises.

This is an escalation. At Chinese state-owned firms it is not unusual for the employer to hold the passports of executives and party officials. What is new is that the state is now extending the same logic into the private sector, and that it selects individuals based on an assessment of how critical they are to the country – not merely on title or employer. The AI engineer has, in effect, been redefined from employee to strategic asset.

From a DeepSeek exception to a broad pattern

The pattern first surfaced in March 2025, when DeepSeek required selected research and development staff to surrender their passports. The stated rationale was to protect commercial and state secrets. What looked at the time like one company's internal measure now appears to have been the start of a broader policy.

By May 2026 the restrictions have widened to several private players. The toolkit combines exit bans, passport confiscation and capital controls. Authorities add names to the list on a rolling basis as they assess who holds knowledge that should not leave the country. The fear Beijing cites is concrete: that researchers could leak sensitive technical information abroad, or that American companies could acquire valuable intellectual property through acquisitions or hiring.

The control does not stop at people. In late April, leading AI firms were reportedly instructed by the powerful planning agency NDRC to reject US-origin capital in upcoming funding rounds unless they first obtain clearance. At the same time, several Chinese AI startups are weighing a move of their corporate registration from overseas back to the mainland, after Beijing blocked a planned acquisition worth around two billion dollars. The money, the people and the corporate structures are to stay within the country's borders.

Why Beijing is tightening its grip now

The timing tracks with the gap to the United States closing fast. Stanford's 2026 AI Index estimates that the performance gap between the best American and Chinese models has narrowed to 2.7 percent, down from somewhere between 17 and 32 percentage points in mid-2023. China now accounts for close to 70 percent of AI patents filed globally, and the flow of Chinese AI talent to the US has dropped sharply since 2017.

Once the lead is gone, the people who built it become the scarce resource. A frontier-model team consists of a small number of individuals, and each of them carries knowledge that took years and enormous sums to build. Beijing treats them accordingly. The logic mirrors American export controls on advanced chips, only inverted: the US tries to stop the hardware from getting out, China tries to stop the minds and the capital from getting out.

What it means for your organization

For leaders this is more than a curiosity from the great-power technology rivalry. Many teams have over the past year leaned toward Chinese models – DeepSeek, Alibaba's Qwen, Moonshot's Kimi – precisely because they are cheap, open and capable. Today's news layers a geopolitical dimension on top of that assessment.

When a state actively confines the people behind a model and dictates which capital the company may take in, it tells you how closely the vendor is tied to state interests. That affects predictability, continuity of supply and the question of where data and dependencies end up. A model can be technically excellent and still carry a vendor risk that does not show up in the benchmark figures.

Concretely, CIOs and CISOs should do three things. Map where Chinese models are actually in use across the organization, including via third-party services and open weights baked into other products. Assess the exposure against a scenario where access, terms or support change for political reasons outside the company's control. And keep at least one genuine alternative vendor warm, so that switching is a decision rather than a crisis.

For the board the point is simpler: AI vendors are no longer neutral technology choices. They sit in the middle of a conflict in which both Washington and Beijing are willing to use coercive measures – export bans on one side, exit bans and capital controls on the other. That risk belongs in the same assessment as other country risk, not in a footnote to the IT budget.

Neither DeepSeek nor the other companies had commented publicly on the expansion when Bloomberg reported it. That in itself is telling. When a country's best engineers cannot freely leave, it is not primarily a personnel matter – it is a signal of how strategic the authorities consider the technology to have become.

Sources and media

• Primary source: Bloomberg, "China Expands Travel Curbs to Top AI Talent at Private Firms", 26 May 2026. source_url: https://www.bloomberg.com/news/articles/2026-05-26/china-expands-travel-curbs-to-top-ai-talent-at-private-firms
• Corroborating coverage: The Japan Times and The Edge Singapore (syndicated from Bloomberg), and The Next Web for details on NDRC capital controls, the expansion to additional firms and the blocked acquisition.
• Context data on the performance gap, patent share and talent migration: Stanford AI Index 2026.
• Thumbnail: OpenAI Image 2 / hogby.ai. No logos, faces or reproduced company UI were used.
• Video skipped this run due to a low credit balance with the image provider.