NeurIPS China Ban: Who Boycotted Whom?

NeurIPS, the world's top machine learning conference, added US sanctions compliance to its 2026 handbook — effectively banning researchers from Huawei, SenseTime, and SMIC. China's science federations called a boycott. Four days later, NeurIPS reversed course and apologised. Reuters headline: "China boycotts top AI conference." SCMP headline: "Top US AI conference apologises." The perception gap on this story hit 6.9 — one of the highest for any tech story this year.

Same facts. Two completely different stories. Here they are, back to back.

Version A: The View From Washington

On March 23, NeurIPS published its 2026 handbook with a new clause. The California-based foundation said it was "required by law to comply with US sanctions and trade restrictions" and could not accept submissions from entities on the Treasury Department's OFAC Specially Designated Nationals list. Standard legal compliance. The kind of thing a US-based nonprofit does when its lawyers tell it to.

China's response was swift and political. The China Computer Federation issued a statement calling on all Chinese researchers to refuse to submit papers, withdraw from reviewer roles, and boycott the conference entirely. The China Association for Science and Technology said it would halt subsidies for researchers attending. Senior Tencent researchers Tu Zhaopeng and Chang Heng resigned their NeurIPS positions.

Reuters ran the story under a clear headline: "China boycotts top AI conference after ban on papers from US-sanctioned entities."

The frame was familiar. A US institution follows the law. China retaliates. The actors are clear: NeurIPS did what it had to do, and Beijing turned a compliance matter into a political confrontation. MarketScreener, Investing.com, and wire pickups across the US all carried the same structure — China as the active boycotter, NeurIPS as the passive rule-follower.

If you read only this version, you'd understand the story as another chapter in China's escalating tech standoff with Washington. Predictable. Almost boring.

Now Flip.

Same week. Same conference. Same reversal. Completely different story.

The South China Morning Post headline: "Top US AI conference apologises after sanctions policy sparks backlash in China."

NeurIPS isn't following the law here. It's admitting a mistake. The conference said it had "gone beyond its legal obligations" and that the original handbook language was the result of "miscommunication between the NeurIPS Foundation and our legal team." They linked to an OFAC database covering 873 Chinese entities — far more than NeurIPS was legally required to block.

Global Times ran: "Top AI conference NeurIPS issues apology for following US sanctions policy after boycott from Chinese academic community." The emphasis falls on the word apology. NeurIPS didn't hold its ground. It buckled.

In this version, the China Computer Federation didn't throw a political tantrum. It defended a principle. The CCF statement called the ban "a politicisation of academic exchange" that "violates fundamental academic principles." The China Society of Image and Graphics said the policy "runs counter to the principles of openness and collaboration in scientific research." Xinhua framed the boycott as Chinese researchers "firmly resisting" discrimination — not starting a fight, but responding to one.

The subject and object switch places entirely. NeurIPS isn't the institution that followed the law. It's the institution that overreached, got caught, and apologised. China's researchers aren't boycotters. They're the reason the conference reversed course.

If you read only this version, you'd understand the story as a victory. Chinese scientific solidarity forced a powerful Western institution to back down. The Pax Silica chip alliance may control who gets computing power — but it can't control who publishes research. Not this week, anyway.

The Ground Shifts

Both versions use real facts. Both are internally consistent. Reuters accurately reports that China's federations announced a boycott. SCMP accurately reports that NeurIPS apologised and admitted it went beyond legal requirements. Neither version is lying.

But one puts China in the subject position — China boycotts — and the other puts the conference there — conference apologises. That single grammatical choice decides who has power, who's reacting, and who won.

Which headline did you see first? And what did it tell you about who's in control of the future of AI research?