Science has always claimed to be above politics. The messy, compromised, flag-waving world of geopolitics wasn’t supposed to touch peer review and conference proceedings. Then NeurIPS — the most important machine learning conference on the planet — spent three chaotic days this week proving that claim was wishful thinking.
Here’s what happened: on March 23, NeurIPS published its Main Track Handbook for 2026 with a new clause that nobody saw coming. Researchers affiliated with entities on the US Treasury Department’s OFAC Specially Designated Nationals (SDN) list could no longer submit papers, serve as reviewers, or hold area chair roles. The affected companies included Huawei, SenseTime, Megvii, Hikvision, and SMIC — all major players in Chinese AI research.
The backlash was immediate, coordinated, and effective. By Friday March 27, NeurIPS reversed course. But the damage — to trust, to the idea of open science, and to US-China research relations — may be considerably harder to undo.
What NeurIPS Actually Did (and Didn’t Do)
The original handbook language wasn’t subtle. It stated that the NeurIPS Foundation cannot provide “services” — explicitly including peer review, editing, and publication — to individuals representing institutions on the OFAC SDN list. Researchers were directed to check the official sanctions database themselves.
NeurIPS framed this as compliance with US law, not a policy choice. Whether that framing holds is another question. IEEE made nearly identical arguments in 2019 when it barred Huawei researchers from peer review, and reversed within weeks under international pressure. History, it turns out, rhymes.
The reversal came with an explanation that only made things worse. NeurIPS said in a statement on X: “In preparing the NeurIPS 2026 handbook, we included a link to a US government sanctions tool that covers a significantly broader set of restrictions than those NeurIPS is actually required to follow. This error was due to miscommunication between the NeurIPS Foundation and our legal team.”
Miscommunication. That’s the word they went with for a policy that nearly fractured one of the world’s most important scientific communities.
China’s Response Was Fast and Organized
Two days after the handbook dropped, the China Computer Federation (CCF) issued a formal statement calling the policy a violation of “openness, inclusiveness, equality and cooperation” — the basic principles of academic exchange. They called on Chinese researchers to withhold services, stop submitting papers, and fully disengage from NeurIPS 2026 unless the policy changed.
Then the China Association for Science and Technology (CAST), a government-affiliated organization, escalated. CAST announced it would stop funding Chinese scholars traveling to attend NeurIPS and redirect that money to conferences that “respect the rights of Chinese scholars.” More pointedly: CAST said it would no longer count NeurIPS publications as academic achievements when evaluating research funding.
That last move is the one that stings. For Chinese researchers, NeurIPS acceptance has been a career credential — a signal to universities and funding bodies that their work meets the highest international standards. Removing it from the approved list doesn’t just inconvenience researchers. It restructures their incentives entirely.
Four named Chinese researchers publicly withdrew from NeurIPS 2026 reviewer and area chair positions before the reversal. The CCF also threatened to remove NeurIPS from its “Recommended International Academic Conferences and Journals” directory — a move that would carry enormous weight across Chinese academia.
Why This Isn’t Just a Conference Drama
China produces a substantial and growing share of the world’s top machine learning research. NeurIPS without Chinese researchers isn’t NeurIPS. It’s a regional conference with a famous name.
Paul Triolo, a partner at the advisory firm DGA-Albright Stonebridge who studies US-China relations, called it “a potential watershed moment.” His argument: attracting Chinese researchers to NeurIPS is actually in US interests, because it maintains communication channels and shared scientific norms. Some American officials want decoupling; others understand that decoupling in basic research doesn’t stay basic for long.
“At some level now it is going to be hard to keep basic AI research out of the political picture,” Triolo told Wired. That’s putting it mildly. The NeurIPS incident isn’t the first — it’s part of a pattern that’s been accelerating since 2019, when US export controls began targeting Chinese semiconductor companies. Every year the line between “civilian AI research” and “dual-use technology” gets harder to draw.
The irony is that this happened at NeurIPS — a conference co-founded by Yoshua Bengio, who has spent decades championing open international scientific exchange. Bengio’s view on AI safety and global collaboration has always been that the risks are too large for any one country to manage alone. The sanctions episode pushed in exactly the opposite direction.
Who Actually Gets Hurt Here?
The companies on the SDN list — Huawei, SenseTime, Megvii — are not going to stop doing AI research because their employees can’t publish at NeurIPS. They’ll publish elsewhere. They’ll build their own conferences. CAST’s threat to redirect funding toward domestic conferences isn’t empty: China has been quietly building an alternative academic infrastructure for years, and moments like this accelerate it.
The people who get hurt are the researchers in the middle. Chinese PhD students at affected companies who were building international careers. American scientists who collaborate with Chinese counterparts and now have to navigate additional bureaucratic uncertainty. The global ML community, which benefits from Chinese research contributions at a rate that’s hard to overstate — the open-source AI wave of early 2026 was substantially driven by Chinese labs.
And there’s a harder question here about what NeurIPS is actually for. If it becomes a conference where participation is contingent on geopolitical alignment, it stops being the global scientific commons it was designed to be. It becomes something more like a trade show for allied nations, with a peer-review veneer.
What Changed (and What Didn’t)
The reversal narrowed the restrictions to the SDN’s Specially Designated Nationals and Blocked Persons list — a much smaller category, used primarily for terrorist groups and criminal organizations. Huawei and SenseTime are not on that list. Practically speaking, the original policy change is gone.
But the CCF and CAST have not fully backed down. CAST’s decision to stop funding Chinese attendance at NeurIPS was not officially reversed as of Friday. The boycott calls were “paused” pending the policy change, but the underlying message was clear: next time, the response will be faster and harder.
This connects to a broader trend we’ve been tracking — from Anthropic’s leaked Claude Mythos and its “unprecedented cybersecurity risks” to Jensen Huang’s claims about AGI. AI is no longer just a technology question. It’s a geopolitical one. The era when researchers could pretend otherwise is ending, conference by conference, policy memo by policy memo.
The cat is out of the bag. (Sorry. Had to.)
Sources
- Wired — AI Research Is Getting Harder to Separate From Geopolitics
- Reuters — Top AI conference reverses ban on papers from US-sanctioned entities after Chinese boycott
- Awesome Agents — NeurIPS Bans Sanctioned Chinese Labs, CCF Calls Boycott
- South China Morning Post — AI rift widens as China urges boycott of top US conference
🐾 Visit the Pudgy Cat Shop for prints and cat-approved goodies, or find our illustrated books on Amazon.
