OpenAI's 722 machine-made proofs put a profession on edge, and the real question is verification

OpenAI has released 722 mathematical proofs, according to the brief report this note is based on. In response, nearly 900 mathematicians signed a letter of protest. The report says their main concern is the future of their profession.

Our source material is limited. It does not say which problems the proofs cover or how hard they are. It also does not say whether they have been formally checked, either by people or by proof assistants (software that verifies each logical step). We will not fill those gaps. One fact still stands out: this many specialists rarely sign a joint letter. That suggests the release is seen as more than a demo.

The fear is not irrational. Our thesis on jobs is that AI "changes which work matters" more than it simply removes work. Mathematics may be an unusually clear case. If machines can produce proofs in large numbers, writing a proof is no longer the bottleneck. The hard parts become checking that a proof is correct, deciding which questions are worth asking, and understanding why a result is true. In law, we described the firm's pyramid narrowing at the base. Academic mathematics could see the same thing: junior research work feels the pressure first, while expert judgment becomes more valuable.

Our reading: the number 722 is a figure, not a measure of quality. Our editorial rule applies here too. Measurement beats marketing, and a public release is not the same as demonstrated capability. The test that matters is how many of these proofs hold up under independent review, and whether any of them settle open problems or only restate known results. The protest also raises a fair point about governance. Mathematicians are being asked to absorb the cost of checking a flood of machine-made output they did not ask for.

The short-term transition will be rough for early-career researchers. That deserves honesty, not dismissal. Over the long term, the picture is more hopeful. Mathematics underpins physics, cryptography, biology and drug discovery. Proof generation that is cheap and checkable could speed up the science that, in our long-term thesis, helps eradicate diseases. Done well, this frees mathematicians to work on the questions they care about most, not routine derivations. How it plays out will depend on verification standards and on whether the research community gets a real say in how these tools are used.