I think we might lose public key cryptography

A Johns Hopkins cryptography professor warns that standardized public-key cryptography could be undermined by rapid advances in machine-led mathematics. He clarifies this is not a call to abandon encryption or an assertion of an imminent breakthrough - no known public cryptanalytic results are pending - but emphasizes that most deployed schemes depend on a small set of hard problems (primarily Module-LWE and ECDLP, with LWE and syndrome decoding as backups). Machines are now outperforming humans on math problems, so the long-held confidence from decades of human analysis may not be sufficient to guarantee current parameter choices.

The practical consequence would be that widely accepted security levels (for example, schemes assumed to provide 128-bit security) might be materially weaker - perhaps 96 or 108 bits - forcing honest standards bodies to deprecate smaller parameter sets. Simply enlarging parameters requires trustworthy margins, and that trust would rest on whether machine progress has truly stalled. Frontier labs are almost certainly exploring cryptanalysis with internal models, so outcomes are uncertain; relying solely on historical human expertise to anchor global Internet security is increasingly risky and may require revising how cryptographic confidence is established.