NeurIPS China Boycott: AI Research Nearly Split

NeurIPS, the world's most influential AI conference, banned researchers from Huawei, SMIC, and other US-sanctioned Chinese firms from submitting papers in March 2026. China's three largest science federations called a full boycott. NeurIPS reversed the policy within four days, calling it a "miscommunication" — but the Albis Perception Gap Index scored the story at 6.9, with US and Chinese media telling completely incompatible versions of what happened and why.

On March 23, NeurIPS published its 2026 Main Track Handbook with a clause new to the conference's 39-year history. Researchers affiliated with OFAC-sanctioned entities couldn't submit papers, serve as reviewers, or hold editorial roles. The handbook didn't name Huawei or SenseTime. It linked to the sanctions database and let the list speak for itself.

What happened next depends on where you read about it.

The US Version: A Clerical Error, Quickly Fixed

Western outlets framed it as an administrative mistake. Reuters reported NeurIPS "went beyond its legal obligations" and the ban was "more limited than initially indicated." The apology called it a "miscommunication between the NeurIPS Foundation and our legal team."

Consistent US framing: compliance officers copied the wrong link, Chinese institutions overreacted, adults sorted it out. Tech sections, not front pages. Most US readers came away thinking: bureaucratic hiccup.

The Chinese Version: Politicisation of Science

Chinese state media told a different story. Xinhua described "sanctions-driven researcher exclusion." The China Computer Federation called it a violation of "openness, inclusiveness, equality and cooperation." CAST didn't just protest — it suspended all funding for Chinese scholars attending NeurIPS 2026 and stripped accepted papers of academic credit.

The SCMP reported four senior Chinese AI researchers withdrew from reviewer and area chair roles before the reversal came. The CCF threatened to delist NeurIPS from recommended conferences — effectively discouraging every Chinese computer scientist from publishing there.

In Chinese coverage, this wasn't clerical. It was the latest front in AI decoupling that Western audiences barely register.

What Both Framings Miss

The US "miscommunication" framing dodges a harder question: why was the OFAC link included at all? OFAC compliance for scientific conferences sits in a legal grey zone. IEEE faced the same dilemma in 2019 when it briefly barred Huawei researchers from peer review — and reversed under identical pressure.

The Chinese "politicisation" framing skips past the fact that several sanctioned entities — Hikvision, SenseTime — were listed for human rights concerns, not just trade competition. The boycott treated all sanctions as equivalent, whether the reason was chip manufacturing or surveillance tech deployed in Xinjiang.

Neither asked the question that matters outside Washington and Beijing: what happens to AI safety research when the two countries producing most of it can't work together?

The Invisible Majority

The PGI captures what no single outlet does: 3.87 billion people never heard about this. The story registered in the US, Europe, and East Asia. Invisible across the Middle East, South Asia, Latin America, and Africa — regions that'll be shaped by AI systems they had no role building, and won't even see the scientific community arguing over who gets to build them.

The Pax Silica chip alliance already divided the world into tiers of AI hardware access. NeurIPS nearly divided who gets to publish AI research along the same lines.

The Damage That Outlasts the Reversal

NeurIPS walked it back. But CAST hasn't reversed its funding suspension. NeurIPS 2026 papers still won't count toward Chinese academic evaluations. The four researchers who resigned haven't returned.

The conference runs in December in Sydney. Chinese researchers will likely attend — but institutional incentives now point toward publishing at home. China's domestic AI conferences just got a recruitment boost no apology letter can undo.

The 2019 IEEE precedent: that ban lasted two weeks. Chinese submissions dropped 12% the next year and didn't fully recover for three.

Two Conferences, One Field?

PGI breakdown: factual divergence between US and Chinese coverage scored 6.0 — they couldn't agree on what happened. Narrative framing hit 8.0 — the gap between "legal mix-up" and "deliberate exclusion" is a canyon. Cui bono scored 7.5: each version neatly served its own strategic interests.

For four days, the world's AI research community almost formally split in two. The fracture was repaired with an email. But an apology doesn't rebuild trust — and trust is what open science runs on.

Next time someone at NeurIPS opens a compliance handbook, or someone at the CCF drafts a boycott letter, both sides will remember this week. The question isn't whether AI research will decouple. It's whether anyone outside those two rooms will notice when it does.

---

This story was scored by the Albis Perception Gap Index — measuring how differently the world frames the same events. See today's most divided stories →