OpenAI announced Monday that its AI models produced three scientific discoveries spanning harmonic analysis, complexity theory, and quantum field theory. The headline result is a claimed proof of the local smoothing conjecture — a central proposition about wave decomposition — in every spatial dimension, extending what human mathematicians had only managed in three dimensions. The company also claims to have disproved the Exponential Time Hypothesis and provided rigorous proof of quantum chromodynamics existence. The verification picture is fractured by design. OpenAI assembled three external expert groups as "shepherds" to translate raw AI output into publishable science. The mathematics and computer science groups chose to release data immediately; the physics group withheld findings to write their own interpretation first. This creates exactly the asymmetry of information OpenAI itself acknowledges — some researchers get early access, others don't, and the company controls who stands in the verification queue. Alex Iosevich, a University of Rochester mathematician whose PhD adviser Chris Sogge formulated the local smoothing conjecture, called the result a "historic development" and expressed both joy and a "twinge of sadness" at what AI-driven proof means for theoretical mathematics. His reaction captures the field's genuine ambivalence: the problems get solved faster, but the human intellectual enterprise that generated them transforms into something unrecognizable. The broader pattern matters more than any single claim. Last month, Anthropic announced an AI-discovered enzyme resembling CRISPR — a finding that drew mixed scientific reception, with some researchers accusing the company of exaggerating significance. OpenAI's $852 billion valuation creates obvious incentives to frame computational outputs as "discoveries" before the verification cycle completes. The shepherd model is better than a raw press release, but it still puts the AI company in control of the narrative timeline. What's genuinely new here is the scope of the claim. Proving a conjecture across all spatial dimensions, if it holds, is categorically different from incremental computational assistance. Disproving the Exponential Time Hypothesis, if validated, would reshape fundamental assumptions in complexity theory about how hard certain problems actually are. These aren't productivity gains — they're assertions about the structure of mathematics itself. The cost structure is worth noting plainly. OpenAI captures the reputational and market value of "AI makes scientific discoveries" immediately. The scientific community bears the verification cost over months or years. If the proofs hold, OpenAI's narrative about artificial general intelligence accelerates. If flaws emerge, the corrections will arrive in journal footnotes that nobody outside the field reads. The twenty-year question isn't whether AI can assist with mathematical proof — it obviously can. It's whether the verification and credit infrastructure can keep pace with a system where the entity generating claims also controls their initial release, selects the first reviewers, and captures the headline value before independent scrutiny is possible.
