How do we prevent mathemathics from devolving into the Medieval Era of secrecy?

This raises alarm that generative AI and massive compute could drive mathematics back toward secrecy by letting wealthy actors brute-force proofs and formalizations before human researchers can publish or interpret them. Historical episodes are invoked - Scipione del Ferro hiding his cubic solution for decades and Newton delaying calculus - to show how secrecy once secured careers. Contemporary worries center on large language models producing mostly correct proofs and fully formalized Lean proofs that are inscrutable to humans, while estimates for corporate-scale solution hunts (e.g., a claimed Navier-Stokes counterexample) range from roughly $6.5 million to $40 million. Terence Tao’s diagnosis is quoted: rumors of promising work can trigger AI-powered flattening that steals problems, disincentivizing open sharing and undermining the research ecosystem.

Proposed remedies include policy rules insisting that only human individuals or teams receive credit for results, and requiring publication of the full methodology used to obtain a result - including prompts and agent setups - to deter corporate secrecy. Another suggestion is to designate classes of problems that demand careful human-centric analysis, not raw solution extraction. An answer reframes AI as accelerating long-running tensions rather than inventing them, recalling the Renaissance shift toward open science (Kepler, Newton, later rigor from Cauchy) and arguing that maintaining incentives for insight, explanation, and community learning is essential to prevent a return to competitive secrecy.