A conference handbook is supposed to tell researchers how to submit a paper. This week, NeurIPS’s handbook helped trigger an international argument over who was welcome to participate.
Chinese scientific organizations objected to restrictions affecting institutional participation. NeurIPS then apologized and clarified its policy, saying it had linked to a sanctions tool broader than the restrictions it was required to apply.
That is serious enough without describing it as a blanket ban on Chinese researchers. Nationality, institutional affiliation and the scope of a sanctions rule are different things. Much of this dispute turns on exactly those distinctions.
What NeurIPS Says Went Wrong
In its public explanation, NeurIPS said the handbook’s link covered a substantially broader range of U.S. restrictions than its own compliance obligations. It attributed the error to communication between the foundation and its legal team, accepted organizational responsibility and apologized.
The organizers said they had corrected the link and clarified the text, bringing it into line with their previous approach and that of other scientific organizations. They emphasized that compliant institutions and individuals remained welcome.
That is the organization’s explanation of the episode. It does not establish that every restriction vanished, or that every company named in discussion belonged to the same sanctions list. Treating several different lists as one universal blacklist is precisely the sort of shortcut that makes a dispute harder to understand.
Two Chinese Organizations, Two Responses
The China Computer Federation’s March 25 statement called for researchers to withhold academic services and submissions. It urged NeurIPS to correct the restrictions and warned that the conference could be removed from CCF’s recommended directory if it did not.
That was a boycott call and a conditional warning about the directory, not proof that every researcher had withdrawn or that removal had already happened.
The China Association for Science and Technology issued a separate March 27 statement. CAST said it would stop accepting applications for funding to attend NeurIPS 2026 and redirect that support toward other conferences.
It also said NeurIPS 2026 papers would not count as representative achievements in applications for CAST-funded projects. Its statement nevertheless allowed recognition of their academic influence following fair evaluation by relevant national societies.
That scope matters. This was not a declaration that a NeurIPS paper had become worthless in every Chinese university or under every funding program. CCF and CAST are also distinct organizations; one statement should not silently become the policy of both.
Why a Handbook Can Have Large Consequences
For researchers, participation is more than a trip and a conference badge. Preparing a submission takes time. Reviewing other people’s work is a commitment. Collaboration depends partly on knowing that the rules will be clear enough to plan around.
Uncertainty creates practical questions: can a team submit together, should a researcher accept a reviewing role, and will travel support still be available? None of those questions improves the paper’s experiments, but all of them can consume the time meant for those experiments.
There is a broader connection to the international exchange behind open AI models. Research travels through shared publications, code and criticism. Making those channels harder to use can affect people well beyond the institutions directly named in a restriction.
A Correction Is a Beginning
As of this March 28 report, NeurIPS has issued its clarification, while CAST’s published response sets out its own funding position. A correction by the conference should not be treated as automatic evidence that every other organization has reversed its response.
The immediate lesson is less dramatic than declaring global science permanently broken, and more useful: explain rules precisely, distinguish obligations from mistakes, and correct misleading guidance quickly.
Trust is harder to edit than a hyperlink. That does not make the correction pointless. It makes getting the next version right rather important.
The cat is out of the bag. Ideally, someone now reads the bag’s terms carefully before announcing which cats were ever excluded.
