What RSEs should take from the OpenAI Navier–Stokes incident
A guide to three posts
In the week of 8 September, OpenAI announced a proof of the forced blowup case of the Navier–Stokes Millennium Prize Problem. I have written a trio of blog posts surrounding what I’ve been calling the OpenAI Navier–Stokes incident.
Most of the arguments I’ve seen, among RSEs too, are about one of these things:
- it isn’t peer reviewed yet, so it isn’t really a proof;
- OpenAI dug into users’ Codex sessions to get it;
- this is mathematics’ AlphaGo moment;
- mathematicians are grieving their craft, and are still in the bargaining phase.
Whichever side you take, each of these pulls attention away from the real issue, and makes it easier for the people who caused it to get away with it.
Unpacking the Navier–Stokes blowup: the mathematics, the machines, and the mathematicians
The foundation: what an RSE needs to know to understand the incident itself, and what not to misread. Among other things, it explains why the validity of the proof isn’t the main concern, why it’s unlikely OpenAI invaded users’ privacy, and why this isn’t an AlphaGo moment. It’s long. If the mathematics isn’t for you, skip to the sections on Lean and ethics.
Progressive disclosure for mathematical proofs
The real issue. AI has changed the cybersecurity community and the mathematics community in parallel ways, and once you see the parallel, the way OpenAI dropped its Navier–Stokes proof looks a lot like disclosing a zero-day security bug irresponsibly. The parallel may sound odd at first. Read it through, and by the end you should see why I call it an incident, and what mathematics could borrow from coordinated disclosure.
Mathematicians aren’t mourning their craft
Where agentic engineering and mathematics part ways. Software engineers have learnt to read the reaction to coding agents as craft versus product. That divide doesn’t apply to mathematics, and this incident has mathematicians reminding each other what they are really after. In the end, I claim research software engineering should be closer to mathematicians in this regard.