AI Research Provenance Disputes

Whether agent-driven “novel research” claims can actually be verified, tracked through three cases in one stretch: Anthropic’s computer-checked Fermat’s Last Theorem formalization (the clean, verifiable case), and OpenAI’s Navier-Stokes and Gromov soficity claims (the messy cases, where credit and training-data-provenance disputes with the mathematicians whose unpublished work was allegedly involved escalated into public accusations of research misconduct).