OpenAI announced on Monday that it is forming an independent advisory group for mathematics, based at the Institute for Advanced Study in Princeton. The group is called the Advisory Group on Mathematics and Artificial Intelligence, and its task is to give mathematicians a voice in the company's research. It sounds good. The problem is that in the same announcement OpenAI wrote in advance what that group must not do.

"The group will not be responsible for advising on how we time our internal progress in mathematics" - that is the wording from the company's blog. In other words: comment all you like on what we have done, but we set the pace. The Princeton institute confirmed it with a sentence worth reading twice: "While we will offer advice, we have no decision-making power at any artificial intelligence company, and responsibility for each company's decisions remains with that company."

When an oversight mechanism explains by itself that it oversees nothing, it is worth asking what it actually controls. The answer, at least according to the powers described, is perception: members will assess the significance of results and coordinate their publication. That is not oversight, it is an editorial desk for press releases - just with better names on the door.

The reason for the whole story is OpenAI's claim that it solved one of the Millennium Problems - Navier-Stokes, a problem mathematics has been waiting on for more than a century - and that the same internal model solved over 100 further open problems. If that survives independent verification, it is a historic moment for mathematics. If it does not, it is the most expensive press release in the history of the discipline. Which is exactly why who does the verifying is not a technical question.

And here the numbers get awkward. The group has nine founding members, and of those only one - Camillo De Lellis, from the Institute itself - signed the open letter in which 25 Fields Medal winners warned that AI labs doing competitive problem-solving threaten intellectual work. The company invited its critics, but invited them at a ratio of one to eight. Members are unpaid, may advise uninvited, may speak publicly and decide for themselves whether to stay. Everything except the one thing that would actually mean something: saying stop.

This is not the first time a Western tech company has assembled the body that will assess it. The Balkans recognises that pattern without translation - commissions appointed by precisely the person they are meant to oversee, with a mandate to write reports that bind nobody. The difference is that here we call it politics and there they call it governance. Does anyone genuinely believe nine unpaid advisers with no veto will slow down something moving on hundreds of billions?