When Science Meets the Front Line: How the US-China Tech War Is Tearing Global AI Research Apart

.

In one week in March 2026, two stories broke simultaneously — and together they signal something historic: the global AI ecosystem is fracturing along geopolitical lines, and there may be no going back.

The World's Most Important AI Conference Becomes a Battlefield

Every December, the world's best artificial intelligence researchers gather at NeurIPS — the Conference on Neural Information Processing Systems — to present their latest breakthroughs, debate ideas, and recruit talent. Founded in 1987, NeurIPS has long been one of the few genuinely global forums where American, Chinese, European, and Asian scientists share a common intellectual space.

That era may now be ending.

On March 23, 2026, NeurIPS published its handbook for the 2026 conference — scheduled for December in Sydney — containing a new clause stating that it could not provide "services," including peer review and publication, to individuals representing entities on the U.S. sanctions list. The rule immediately affected major Chinese technology firms including Huawei, SenseTime, Megvii, and Hikvision — all of which have researchers who regularly present cutting-edge AI work at the conference.

The backlash was immediate and ferocious.

The China Computer Federation — one of China's most influential professional bodies for computer scientists — issued a statement "strongly opposing" the decision, declaring that "NeurIPS's ban on submissions from specific institutions and its politicisation of academic exchange violate the basic principles of openness, inclusiveness, equality and cooperation that the international academic community has upheld for centuries."

Within days, China's three most powerful scientific organizations had joined the backlash. The China Association for Science and Technology announced the immediate suspension of funding for Chinese scholars wishing to attend NeurIPS, barred NeurIPS-accepted papers from qualifying for its grant programs, and redirected researchers toward domestic conferences or "international conferences that respect the rights and interests of Chinese academics." The China Computer Federation and the Chinese Association of Automation urged Chinese computer scientists to refuse submitting papers to, or providing peer review services for, the conference entirely.

An Apology — But the Damage Was Done

Faced with the tsunami of criticism, NeurIPS quickly retreated — partially.

The conference's organizers issued an apology, acknowledging that their handbook had "included a link to a U.S. government sanctions tool that covers a significantly broader set of restrictions than those NeurIPS is actually required to follow," and attributed the error to "miscommunication between the NeurIPS Foundation and our legal team."

The handbook was updated to specify that the actual restrictions apply only to the Specially Designated Nationals and Blocked Persons list — a narrower category used primarily for terrorist organizations and criminal networks — rather than the far broader export control lists that had originally been cited, which would have swept in thousands of Chinese university and corporate researchers.

But the retreat satisfied almost no one. China's scientific bodies refused to stand down. China's national science association condemned NeurIPS for "introducing political hegemony into academic exchanges, blatantly trampling on the cooperative rules that the academic community has adhered to for hundreds of years, seriously distorting academic fairness, and polluting the academic ecosystem."

"This is a potential watershed moment," said Paul Triolo, a partner at DGA-Albright Stonebridge who studies US-China relations. He argued that attracting Chinese researchers to NeurIPS is beneficial to U.S. interests — but acknowledged that some American officials have been pushing actively for American and Chinese scientists to decouple their work, especially in AI.

While Scientists Argued — Engineers Were Being Detained

The NeurIPS controversy erupted in the same week as an even more dramatic story — one that revealed just how far Beijing is prepared to go to assert control over its AI talent and technology.

On March 25, the Financial Times and Reuters reported that Chinese authorities had imposed exit bans on two co-founders of Manus AI — CEO Xiao Hong and Chief Scientist Ji Yichao — barring them from leaving mainland China while regulators reviewed Meta's $2 billion acquisition of the company. The two executives, who had been living and working in Singapore, were summoned to Beijing for questioning sessions focused on how Manus had restructured itself prior to the Meta deal. They were told they could not leave.

Manus is not a household name outside the tech world — but inside it, the company had become a sensation. Dubbed "China's next DeepSeek," Manus had built what it called the world's first fully autonomous AI agent — a system capable of buying property, programming video games, analyzing stocks, and planning travel itineraries without human oversight. It reached $100 million in annualized recurring revenue within just eight months of launch.

Meta had announced the acquisition in December 2024, valuing the company at approximately $2 billion and framing it as a cornerstone of its new "Meta Superintelligence Labs" division. To reassure U.S. regulators, Manus had carefully unwound its Chinese ties — relocating its headquarters from Shenzhen to Singapore, replacing state-linked Chinese investors with U.S. venture capital, and shutting down its entire China-based engineering team.

Beijing was not impressed by the restructuring. In fact, it was the very act of restructuring that angered authorities most.

The "Singapore Bath" — and Why Beijing Is Calling It Out

Central to China's investigation is what regulators describe as a deliberate effort by Manus to escape Chinese oversight. Authorities are examining whether the core AI agent intellectual property — developed in Shenzhen — was moved to the Singapore entity without the required government approvals under China's Regulations on Technology Import and Export Administration.

The Manus case has exposed the limits of a strategy that Chinese tech entrepreneurs had come to rely on — known informally in industry circles as the "Singapore bath": build technology in China, reincorporate offshore to facilitate foreign investment and exit, then present yourself as a Singapore-based company for the purposes of a Western acquisition. Beijing's willingness to pierce the corporate veil based on the underlying technology's origin and the founders' Chinese ties transforms offshore restructuring from a compliance strategy into a compliance vulnerability.

This marks the first time Beijing has used exit bans on executives to directly impede a multibillion-dollar deal with a U.S. technology firm — a significant escalation that moves beyond administrative review into personal coercion.

For Meta, the human capital risk is acute. The value of Manus is intrinsically tied to its founders and key engineers. Blocking their movement creates immediate friction for integration and raises serious questions about Meta's ability to fully realize the strategic benefits of the acquisition. For Chinese AI founders generally, the precedent is chilling: the promise of a clean break from Chinese regulation is now in serious doubt.

Two Tracks, One Direction: Decoupling

Taken together, the NeurIPS controversy and the Manus exit bans are not isolated incidents. They are symptoms of a structural transformation in how the world's two AI superpowers relate to each other — and to the global scientific community caught between them.

Washington has in recent years dramatically increased scrutiny of Chinese scientists at U.S. universities, investigating a growing number over alleged ties to entities in mainland China, and has imposed sanctions on hundreds of Chinese universities and companies — preventing them from legally acquiring a wide range of advanced U.S. technology. China has responded by tightening its own regulatory grip on outbound technology flows, using tools ranging from export control reviews to personal travel restrictions.

This tit-for-tat climate complicates collaboration in a sector that both nations are racing to lead. NeurIPS's ban on sanctioned entities could reshape AI research flows, potentially isolating major Chinese players from the world's most prestigious AI platform — while simultaneously accelerating China's investment in domestic alternatives.

CAST has already signaled the direction of travel, saying it will redirect funding toward domestic AI conferences and "international conferences that respect the rights and interests of Chinese academics" — essentially beginning the construction of a parallel scientific ecosystem, insulated from U.S. sanctions law and oriented around Chinese institutions.

The implications extend far beyond conference attendance lists. AI is a field that has historically advanced through open, global collaboration — researchers building on each other's published work, sharing datasets, jointly developing benchmarks. If that collaboration fractures along national lines, the pace of progress slows for everyone, while the risks of parallel and incompatible AI development increase.

A World Dividing Itself in Real Time

There is something both historically significant and quietly tragic about what is unfolding. The internet was built on the premise of borderless information exchange. Modern science was built on the premise that knowledge belongs to humanity, not to states. NeurIPS was built on the premise that a researcher from Beijing and a researcher from Boston could share a stage, a peer review process, and a community of inquiry.

All of those premises are now under pressure — not because scientists want it that way, but because governments have decided that artificial intelligence is too strategically important to leave to scientists alone.

As Paul Triolo observed, AI research is getting harder to separate from geopolitics. The question now is not whether decoupling will happen — it already is — but how far it will go, how fast, and what will be lost in the process.

The lab coats are on the front line now. And nobody asked them if they wanted to be there.

.

Sources:

1. Caixin Global – "Top AI Conference Tightens Sanctions Compliance, Snaring Chinese Tech Giants" (March 26, 2026): https://www.caixinglobal.com/2026-03-26/top-ai-conference-tightens-sanctions-compliance-snaring-chinese-tech-giants-102427750.html
2. South China Morning Post – "AI Rift Widens as China Urges Boycott of Top US Conference Over Sanctions Ban" (March 26, 2026): https://www.scmp.com/tech/article/3348006/ai-rift-widens-china-urges-boycott-top-us-conference-over-sanctions-ban
3. South China Morning Post – "Top US AI Conference Apologises After Sanctions Policy Sparks Backlash in China" (March 27, 2026): https://www.scmp.com/tech/article/3348199/top-us-ai-conference-apologises-after-sanctions-policy-sparks-backlash-china
4. Xinhua – "Chinese Scientific Groups Boycott AI Conference Over Sanctions-Driven Researcher Exclusion" (March 27, 2026): https://english.news.cn/20260327/ca409f62b8ae42e686cdb9e0064b57f6/c.html
5. Washington Post – "China Bars Executives at Meta-Owned AI Company From Leaving Country" (March 25, 2026): https://www.washingtonpost.com/national-security/2026/03/25/meta-manus-china-executives-banned/
6. Euronews – "China Bans Manus Founders From Leaving Country After Meta Acquires AI Startup" (March 26, 2026): https://www.euronews.com/next/2026/03/26/china-bans-manus-founders-from-leaving-country-after-meta-acquires-ai-startup-and-reviews
7. The Legal Wire – "China Restricts Manus Founders From Leaving the Country Amid Investigation Into Meta's $2bn Acquisition": https://thelegalwire.ai/china-restricts-manus-founders-from-leaving-the-country-amid-investigation-into-metas-2bn-acquisition/
8. The Decoder – "Meta Pays $3 Billion for Manus AI After Startup Cut All Chinese Ties to Clear Regulatory Hurdles" (December 31, 2025): https://the-decoder.com/meta-pays-3-billion-for-manus-ai-after-startup-cut-all-chinese-ties-to-clear-regulatory-hurdles/
9. eWeek – "China Blocks Manus Founders, Meta AI Deal" (March 26, 2026): https://www.eweek.com/news/china-blocks-manus-founders-meta-ai-deal/
10. Wired / Overpasses for America – "AI Research Is Getting Harder to Separate From Geopolitics" (March 27, 2026): https://overpassesforamerica.com/ai-research-is-getting-harder-to-separate-from-geopolitics/

.