China Tightens Overseas Travel Review for Top AI Talent Amid U.S.-China Tech Competition

First, the conclusion:

This is not about “controlling all programmers,” but rather about precisely targeting a very small number of top AI and semiconductor talent. The policy focuses on the cross-border transfer of critical knowledge, core expertise, and strategic technological capabilities.

A Bloomberg report on May 26 brought to the forefront an issue that had long been simmering: amid escalating U.S.-China competition in AI, China is reportedly tightening controls on overseas travel for top AI talent. According to Reuters, which cited the Bloomberg report, these measures primarily target leading professionals at companies such as Alibaba and DeepSeek who work on advanced, strategic AI projects, requiring some to obtain approval from relevant authorities before traveling abroad. However, Reuters specifically noted that it has not been able to independently verify the report.

Therefore, the most accurate statement is not that "China has implemented a unified control policy for AI talent exiting the country," but rather that foreign media report China is increasing scrutiny on the overseas travel of a small number of top AI professionals; however, the full public implementation details have not yet been disclosed.

Based on the disclosed information, the policy is truly focused not on ordinary application developers or programmers in the general sense, but on a very small number of individuals who control critical technological pathways: core researchers of large models, heads of model training, AI chip architects, personnel involved in advanced processes or key EDA toolchains, and top technical managers capable of integrating algorithms, computing power, data, and engineering systems.

The policy does not regulate labor mobility itself, but rather the cross-border flow of key knowledge carriers.

A person can certainly move freely. But when that person carries with them a comprehensive set of expertise—including model architectures, training methods, distributed computing resource scheduling, chip design, inference optimization, data governance, and engineering tuning—their mobility is no longer merely a contractual issue, but may fall under the intersecting domains of national security, export controls, confidentiality management, and technological competition.

Over the past few years, the central focus of the U.S.-China AI competition has been chips. The U.S. Department of Commerce’s Bureau of Industry and Security (BIS) has continuously restricted China’s access to advanced AI chips and related manufacturing capabilities through tools such as the Export Administration Regulations (EAR), the Entity List, controls on advanced computing chips, and restrictions on semiconductor manufacturing equipment.

On October 7, 2022, the BIS introduced new restrictions on advanced computing chips, end uses of supercomputers, and semiconductor manufacturing-related items, and expanded the application of the foreign direct product rule to certain existing entities on the Chinese entity list. On October 17, 2023, the United States further strengthened export controls related to advanced computing chips and semiconductor manufacturing equipment based on the 2022 rules.

The core of this logic has never been just the "chip" itself. The chip is merely the surface; what is truly controlled is the underlying computational advantage, training capability, military potential, and control over the next-generation technological infrastructure between nations.

Therefore, once hardware is regulated to a certain extent, policy naturally progresses to the second layer: the flow of knowledge, experience, and talent. This is why the review of AI talent leaving the country cannot be simply understood as traditional "personnel management."

It is more like the spillover of export control logic into personnel scenarios: When technology has become deeply internalized within a small number of core researchers and engineering leads, the cross-border movement of people may be viewed as the transfer of technological capability across borders.

Relying solely on the Export Control Law to explain this issue may be an overstatement, but completely ignoring the Export Control Law would miss the key logic. Such policies may simultaneously rest on three sets of legal tools.

The first group is export control regulations.

Article 2 of the Export Control Law states that controlled items include goods, technology, services, and related technical documents and other data; export control not only encompasses the transfer of controlled items from within China to overseas destinations, but also includes Chinese citizens, legal entities, and non-legal organizations providing controlled items to foreign organizations or individuals.

The Regulations on Export Control of Dual-Use Items (hereinafter referred to as the "Regulations"), announced in 2024 and effective from December 1, 2024, further refine this framework. Article 2 of the Regulations explicitly states that dual-use items include goods, technologies, and services, as well as related technical documentation and other data; the export control context also covers commercial exports, foreign gifting, exhibitions, cooperation, assistance, and other forms of transfer.

This means that in scenarios such as technology transfer abroad, technological collaboration, cross-border R&D, overseas employment, and overseas consulting services, if controlled or sensitive technical information is involved, enterprises and individuals cannot simply view the issue as “has the person left?”—they must ask: has knowledge, technology, data, or services been transferred abroad in any form?

The second group is the entry and exit management rules.

Article 12 of the Exit and Entry Administration Law specifies several circumstances under which Chinese citizens are prohibited from exiting the country, including cases where “there is a potential threat to national security or interests, as determined by the relevant competent departments of the State Council.”

This is the more direct legal provision addressing the issue of "whether one can exit the country." Therefore, if future restrictions or approvals are imposed on specific key AI personnel regarding exit from the country, the legal framework may not be entirely based on export control laws, but could also be implemented through pathways such as national security, entry and exit management, or approval by relevant authorities.

Group three is the confidentiality management rules.

The State Secrecy Law implements classified management of personnel handling state secrets and stipulates that such personnel must obtain approval from relevant authorities before traveling abroad; if the relevant authorities determine that their departure would endanger national security or cause significant harm to national interests, approval for departure shall be denied. After leaving their positions, personnel handling state secrets may still enter a confidentiality period during which they are prohibited from violating regulations regarding employment and travel abroad.

The 2024 revised Regulations for the Implementation of the Law on Guarding State Secrets further stipulate that the departure of personnel handling state secrets shall be subject to recommendations from the personnel and organizational departments and the confidentiality work institutions of the respective agencies or units, and shall be approved in accordance with established personnel and foreign affairs approval authorities; personnel handling state secrets must undergo confidentiality training prior to departure and promptly report any relevant circumstances while abroad.

The lesson for the AI industry is: Not all core technologies are inherently state secrets, but once certain projects, data, model capabilities, chip designs, or dual-use applications are brought under confidentiality or national security frameworks, personnel mobility shifts from a standard employment issue to a security review matter.

Over the past three years, the key terms in AI competition were H100, A100, HBM, advanced packaging, EUV, and EDA. But now, a new keyword is emerging: people.

The reasons are as follows:

First, the marginal effectiveness of hardware controls is being challenged.

U.S. restrictions on advanced chips and semiconductor equipment have indeed significantly increased the difficulty for Chinese companies to acquire high-end computing power. However, at the same time, Chinese companies are accelerating their search for alternative solutions. Reuters has reported that Huawei's Ascend 910B is considered one of the most competitive non-NVIDIA AI chips in the Chinese market; in 2025, Huawei plans to大规模 ship the Ascend 910C to Chinese customers as part of efforts by Chinese AI companies to find alternatives to NVIDIA.

This does not mean China has solved all advanced chip challenges. On the contrary, manufacturing yield, advanced processes, ecosystem compatibility, software stack migration, and cluster stability remain hard constraints. But it shows that merely restricting hardware may not be sufficient to permanently lock down all capabilities.

Second, top talent is harder to replace than hardware.

Chips can be bought, stockpiled, or gradually replenished through alternative routes. But the number of people who can truly organize the training of千亿-parameter models, handle large-scale distributed training failures, design AI chip architectures, and enable coordinated optimization between models and chips is extremely small.

The most scarce resource in the AI industry is not people who can write code, but those who know how to align code, computing power, data, models, engineering systems, and business scenarios into a cohesive force.

Third, once knowledge is transferred, it is difficult to reverse.

Chips have logistics trails, devices have customs records, software has licensing boundaries, and data has transmission traces. But experience is different. When a core researcher joins an overseas team, what they take may not be a specific file, but an entire set of judgment: which paths won’t work, which parameter combinations are effective, which engineering pitfalls are most deadly, and which system bottlenecks deserve the highest priority.

This type of knowledge is difficult to track and nearly impossible to "recover." Therefore, for governments, pre-emptive review is more effective than post-hoc accountability; for businesses, compliance with employee knowledge will become increasingly important.

This type of policy is most easily misinterpreted as two extremes.

One claim is that "Chinese tech talent can no longer leave the country at all." This is clearly an exaggeration. Another claim is that "this is merely media sensationalism and has nothing to do with corporate compliance." This is too dismissive. A more prudent assessment is: If the reports are accurate, the impact will not be evenly distributed among all technical staff, but will be concentrated on a very small number of critical roles.

For ordinary programmers, application developers, and general algorithm engineers, publicly available information does not support the conclusion that "general outbound travel is restricted." Normal business trips, conference attendance, studying abroad, job changes, and overseas employment still primarily follow standard出入境 regulations, employment contracts, non-compete restrictions, confidentiality obligations, data compliance rules, and export control regulations.

However, the situation may differ for core researchers of large models, chip architects, training platform leads, and key technology executives. Whether it’s long-term assignment overseas, short-term participation in sensitive technology conferences, or being acquired, poached, or assigned to cross-border R&D collaboration by overseas companies, these scenarios may trigger higher-level internal reviews and external compliance assessments.

For Chinese tech companies with overseas R&D centers, the impact is more immediate. In the past, companies primarily considered visas, compensation, taxes, employment contracts, and non-compete restrictions when relocating core technical staff. Now, they must ask additional questions: Has this person had access to controlled technologies? Do they possess knowledge of unpublished model capabilities? Are they involved with sensitive data? Have they participated in national key projects? Do they hold classified or quasi-classified positions? Is internal approval, confidentiality review, or export control assessment required?

This is not an issue that HR can resolve alone, but rather one that requires collaboration among legal, compliance, business, security, data governance, and human resources teams.

It is not accurate to simply say “yes” or “no.” Based on existing public regulations, the review of AI talent exiting the country does not automatically prevent Chinese tech companies from listing in the U.S. The core review of listings still focuses on corporate governance, financial disclosures, VIE structure, data security, cybersecurity reviews, overseas listing filings, U.S. securities regulation, and audit working papers.

In 2023, the China Securities Regulatory Commission issued the "Interim Measures for the Administration of Overseas Securities Issuance and Listing by Domestic Enterprises," bringing domestic enterprises' overseas listings under a filing-based regulatory framework; Article 7 of the "Cybersecurity Review Measures" also explicitly states that network platform operators handling personal information of over one million users must apply for a cybersecurity review before listing abroad. This demonstrates that the review logic for technology companies seeking overseas listings has long extended beyond financial and securities law considerations, inherently incorporating assessments of national security, data security, and cross-border regulatory cooperation.

If the review of AI talent leaving the country becomes further institutionalized, it may not directly change the answer to “whether listing is possible,” but it will increase the cross-border compliance costs for certain companies. If an AI company heavily relies on a small number of core technical personnel who are involved in sensitive models, chips, training data, or dual-use technologies, then due diligence by overseas investors, management roadshows, establishment of overseas R&D teams, overseas M&A and restructuring, and IP relocation arrangements may all require more detailed compliance justifications.

The issue is not whether listing is possible, but whether the people, technology, data, IP, and overseas communications involved in the listing process can withstand scrutiny from the perspectives of national security and export controls.

This round of policy signals has brought to the forefront an issue that was often overlooked in the past: the compliance obligations of technology companies now extend beyond data, code, chips, equipment, and contracts to include knowledge in the human mind. In the past, technology compliance most commonly involved three categories: data compliance, export compliance, and intellectual property compliance.

Now, a fourth category of compliance is emerging: personnel knowledge compliance.

First, identify key positions. Companies need to know which roles genuinely have access to core models, chip design, training systems, critical data, and unpublished technical pathways.

Second, establish tiered personnel classification. Not all engineers require the same level of management, but core researchers, architects, training leads, chip design leads, and technical executives should have clearer categorization.

Third, control knowledge access. Who can view the core code? Who can access the training data? Who can download model weights? Who can access experiment logs and critical failure records? These cannot rely solely on default trust.

Fourth, standardize cross-border collaboration. All overseas teams, foreign consultants, overseas laboratories, joint R&D, international conferences, paper publications, and open-source projects must be evaluated for potential involvement in technology exports, data transfers abroad, or leakage of trade secrets.

Fifth, design offboarding and overseas departure procedures. When key personnel leave the company, take up positions abroad, join competitors, or participate in overseas startup projects, they should not only go through employment contract procedures but also trigger simultaneous reviews of confidentiality, non-compete, export control, and data security assessments.

Ultimately, as the AI industry moves into deeper waters, what truly holds value is becoming less like traditional assets. It may not be a machine, not a document, not a set of code, but rather the experience in the minds of a few—knowing how to do something and why it’s done that way.

This is the essence of the "era of targeting the elite." It's not about controlling everyone, but about controlling the most critical flows of knowledge.

[1] Reuters, “China restricts overseas travel for top AI talent at Alibaba and DeepSeek, Bloomberg News reports,” May 26, 2026.

https://www.reuters.com/world/asia-pacific/china-restricts-overseas-travel-top-ai-talent-alibaba-deepseek-bloomberg-news-2026-05-26/

[2] The People's Republic of China Export Control Law, adopted on October 17, 2020, effective December 1, 2020, official text published by the Ministry of Commerce.

https://www.mofcom.gov.cn/zfxxgk/gkml/art/2020/art_76b5fa416a4c42afa8af295eefffdffc.html

[3] Regulations of the People's Republic of China on Export Control of Dual-Use Items, State Council Order No. 792, published on September 30, 2024, effective December 1, 2024, official text from the Ministry of Commerce.

https://www.mofcom.gov.cn/zcfb/dwmygl/art/2024/art_6ed1d5c0336843c493bd45a1911eab17.html

[4] The People's Republic of China Exit and Entry Administration Law, public text from the National Immigration Administration.

https://www.nia.gov.cn/n741440/n741547/c1013311/content.html

[5] The People's Republic of China State Secrets Law, 2024 Revision, Public Legal Text.

https://fgk.chinatax.gov.cn/zcfgk/c100009/c5211806/content.html

[6] Implementation Regulations of the People's Republic of China State Secrets Protection Law, 2024 Revision, republished by the Supreme People's Procuratorate.

https://www.spp.gov.cn/spp/fl/202407/t20240722_660950.shtml

[7] U.S. Federal Register, BIS, “Implementation of Additional Export Controls: Certain Advanced Computing and Semiconductor Manufacturing Items; Supercomputer and Semiconductor End Use; Entity List Modification,” effective October 7, 2022.

https://www.federalregister.gov/documents/2022/10/13/2022-21658/implementation-of-additional-export-controls-certain-advanced-computing-and-semiconductor

[8] U.S. Department of Commerce, BIS, “Commerce Releases Clarifications of Export Control Rules to Restrict the PRC’s Access to Advanced Computing and Semiconductor Manufacturing Items,” April 4, 2024.

https://www.bis.gov/press-release/commerce-releases-clarifications-export-control-rules-restrict-prcs-access-advanced-computing

[9] eCFR, Export Administration Regulations, 15 CFR Parts 730–774.

https://www.ecfr.gov/current/title-15/subtitle-B/chapter-VII/subchapter-C

[10] eCFR, Entity List, Supplement No. 4 to Part 744.

https://www.ecfr.gov/current/title-15/subtitle-B/chapter-VII/subchapter-C/part-744/appendix-Supplement%20No.%204%20to%20Part%20744

[11] Reuters, “AI chip demand forces Huawei to slow smartphone production,” February 5, 2024.

https://www.reuters.com/technology/ai-chip-demand-forces-huawei-slow-smartphone-production-sources-2024-02-05/

[12] Reuters, “Huawei readies new AI chip for mass shipment as China seeks Nvidia alternatives,” April 21, 2025.

Huawei readies new AI chip mass shipment as China seeks Nvidia alternatives, sources say

[13] Q&A on the Cybersecurity Review Measures, National Internet Information Office, January 4, 2022.

https://www.cac.gov.cn/2022-01/04/c_1642894602460572.htm