In a domain where truth admits no approximation, OpenAI quietly stepped back from a Caltech mathematics hackathon this week after a coalition of mathematicians issued an open letter challenging the reliability of AI-generated mathematical reasoning. The withdrawal, unaccompanied by public explanation, arrived at a moment when the gap between what AI systems are marketed to do and what they can actually demonstrate is drawing sustained scrutiny from those best positioned to judge. It is a small but telling episode in the longer story of how institutions learn to calibrate trust in new technolog