Mathematicians Call for OpenAI Boycott After 700 AI-Written Proofs Land at Once

OpenAI posted more than 700 AI-written mathematical manuscripts at the same time. A day later, the company withdrew three of them: the proofs contained a sign error. The Association for Human Mathematics responded with a call to boycott OpenAI. Fields Medal winner Terence Tao said that mass "harvesting" of open problems could leave entire branches of mathematics less fertile.

What happened

Mathematics moves slowly. Colleagues read a paper first, then reviewers, and only after that does it earn a reputation as verified. OpenAI broke that rhythm: hundreds of texts appeared at once, and humans have to check them.

The sign error in three papers showed where the bottleneck is. Generating a plausible proof is fast. Confirming that it is correct means checking every step, and that takes person-hours. The source materials do not say how these particular three errors ended up in the release or how the remaining papers were checked.

The association's name speaks for itself: it advocates for mathematics done by humans. In its wording, the boycott targets OpenAI as a company, not AI tools in general.

Tao's argument: soil depletion

Tao's position is more interesting than simple worry about quality. He is not saying that AI makes mistakes. His concern is that AI succeeds too broadly. Open problems in mathematics are a finite resource. Each one is a point of growth: around a hard problem, methods, schools and entire subfields take shape over years. If a system is run on everything and "picks off" solutions to the problems it can reach, young researchers lose the material they learn from and build careers on.

Tao describes this with the word harvesting. If problems are cleared from the field without growing new theory around them, bare ground is what remains. A solution may be correct, but the understanding that usually comes with it does not accrue to the community.

Key facts of the story:

• Volume: more than 700 manuscripts published at once.
• Withdrawal: three papers withdrawn the next day.
• Reason for withdrawal: a sign error.
• Community reaction: the Association for Human Mathematics announced a call to boycott OpenAI.
• Tao's main argument: mass solving of open problems depletes the "fertility" of branches of mathematics.

How journals and researchers can handle this

The story raises concrete practical questions that now have to be answered.

First: formal verification. If a proof is machine-generated, it is reasonable to require a checkable formal artifact alongside it, not just a PDF. In such a system, a sign error is caught automatically rather than through a reviewer's effort.

Second: provenance labeling. Journals and archives need clear rules: who the author is, how much of the work the model did, and who is responsible for errors. Without them, a withdrawn paper becomes a reputational problem for the entire field.

Third: limits on batch publication. No editorial process can review seven hundred texts at once. The community will have to agree on a pace that matches its capacity to verify.

Fourth, for researchers themselves. Those who work on open problems should remember that a problem's value is not limited to its solution. The theory around it remains in human hands, and that is something AI does not yet replace.

The boycott may turn out to be a symbolic gesture. But the question behind it is not symbolic: who checks mathematics, and how, when writing it becomes cheaper than reading it.