OpenAI's Advisory Board — why consultation after the fact cannot fix what structure broke

OpenAI has made genuine breakthroughs in mathematics. It has also repeatedly announced them in ways that felt like theft to the people who saw them coming — work described, explained, claimed without the mathematicians whose ideas shaped it being told first, or at all. The latest breach, the advisory board, is supposed to fix this. It will not, because it is solving the wrong problem.

When computer science separated from mathematics as its own discipline in the 1960s and 70s, the split was not a disagreement — it was a takeover. Computational mathematics was being redefined and controlled by computer scientists. Mathematicians lost institutional voice. Universities added mathematician seats to computer science advisory boards, genuinely believing this would restore balance. It did not. The boards had no veto power. They had no control of hiring, curriculum, or research direction. They had input after decisions were already made.

OpenAI's new advisory board is structurally identical. Mathematicians are being invited to comment after OpenAI has already chosen what to build, what to announce, and how. The difference between having a voice in a system and having influence over it is not subtle. It is the entire architecture of power. A mathematician on an advisory board can say "this will alienate people." She cannot actually prevent the announcement. She cannot control what gets built next.

The pattern compounds because it is invisible as a pattern. Each breach looks like a mistake. A communication problem, a coordination failure. It is not. It is what happens when one institution has the resources to move fast and the other has only the ability to complain after. OpenAI could redesign this: invite mathematicians into the decision before the work is public, give them veto power on what gets announced and when. That would require surrendering some speed. Instead, it chose the retrofit. The advisory board, the appearance of consultation, the structure that failed fifty years ago and will fail again.

If you have ever joined a committee to influence something that had already been decided, you know the exact feeling: acknowledged but powerless. That is not consultation. It is an institutional design for making power feel shared while keeping it concentrated. OpenAI did not invent this move. But they chose to repeat it, and mathematicians should assume they will keep choosing it until the actual structure changes.