You're just moving the problem around. Instead of OpenAI and Anthropic dumping proofs, you'd instead have someone (maybe an aspiring graduate student, maybe a hobbyist, maybe an established mathematician who has disdain for existing process) dump these proofs en masse personally.