When OpenAI published hundreds of AI-produced mathematics manuscripts on October 6, Alex Wissner-Gross saw one gap in the list of subjects. "Did you pick up on what form of math was missing?" Wissner-Gross, a computer scientist and founder of Reified, asked his co-hosts on Moonshots with Peter Diamandis. Peter Diamandis, the founder of XPRIZE, answered right away: "Encryption."
Wissner-Gross said cryptography was the one category with no results, and he called that "very suggestive." The episode was recorded on October 13, 2026, and published two days later. It became an argument about whether AI that can do mathematics at scale threatens the codes behind bank transfers, cryptocurrencies and secure messages. The hosts did not settle it.
Why hard math matters to encryption
Modern encryption depends on certain mathematical problems taking far too long to solve. A hash function, for example, scrambles data in a way that is easy to compute and is supposed to be practically impossible to reverse. Bitcoin's "proof of work" depends on that property. Computers compete to find inputs that produce a hash with a particular pattern, and the only known way to do it is enormous trial and error.
Wissner-Gross put the dependence plainly. "Crypto depends on math being hard, at least proof of work, but arguably all of crypto depends on math being hard," he said. In many cases, he added, it also depends on the math staying hard in the future. He said cryptocurrency was where the release would probably have its most direct and immediate impact ("not investment advice"), "because this has demonstrated that math is no longer hard. Math can be scalably solved."
Diamandis said the stakes went beyond crypto: "It's your bank account. It's nuclear codes. It's everything." Wissner-Gross answered "yes and no" and took the cases one at a time. If someone steals a bank balance, he said, there are "non-mathematical social remedies": the police can go after the thief, and the bank can correct the entry in its central database. If someone steals a cryptocurrency cold wallet, which is offline storage for the keys that control the funds, the theft is "really hard to fix." As for nuclear codes, he said he would like to think they are kept sufficiently offline, with backstops such as red phones.
Why was cryptography left out?
Diamandis asked whether encryption was left out on purpose. Wissner-Gross said it was "not because it's hard." His presumption was that cryptography is "export controlled in many cases," so including it would have been a problem. That was his guess, not an explanation from OpenAI. The released material, which OpenAI keeps in a public repository, comes with the company's own warning that verification is incomplete and that results not checked by formal proof software may contain errors.
Wissner-Gross also brought up Strawberry and Q*, project names linked to OpenAI's early work on reasoning models, which are models trained to work through problems step by step before answering. He presented the story as allegations only. He said they hold that the task behind OpenAI's discovery of reasoning models was an internal attempt to invert sophisticated, cryptographically hard hash functions. If those allegations were "ultimately proven correct," he said, it would be ironic that the culmination of reasoning models, "basically cooking math," left cryptography out.
"It's AI solving math"
The subject came back in the listener questions. A listener asked: with math "being cooked," what are the chances of a breakthrough that destroys modern encryption? Wissner-Gross answered "exceptionally high." As evidence he again cited the unproven Strawberry and Q* allegations and the missing cryptography category, which he called "exhibit B."
Then he offered what he called a "hotter take." Post-quantum cryptography, he said, would not be "materially safer" than conventional cryptography. Post-quantum cryptography is the new generation of codes designed to survive attacks by future quantum computers. "All these folks in the crypto community who are so concerned about quantum attacks against blockchains, completely misdirected," he said. "It's not quantum that's going to disrupt you. It's AI solving math."
The standards he was targeting are real and recent. In August 2024 the U.S. National Institute of Standards and Technology (NIST) approved three post-quantum standards. ML-KEM handles key establishment, which lets two parties agree on a shared secret. ML-DSA and SLH-DSA are digital-signature schemes that verify who signed something and help detect altered data. They grew out of algorithms called CRYSTALS-Kyber, CRYSTALS-Dilithium and SPHINCS+, and they use more than one mathematical approach. NIST chose them to resist future quantum attacks. Wissner-Gross's point was that a different threat, AI finding mathematical weaknesses, was not the one they were picked to withstand. He gave no example of any of these systems, or of any other cryptosystem, being broken.
An arms race, but which side is ahead?
Salim Ismail, founder of Open ExO and a general partner at Exponential Venture Capital, pushed back twice. In the first exchange he argued that Bitcoin is "not just mathematics founded." Its real breakthrough, he said, was solving the Byzantine generals problem: how many parties who do not trust one another can agree on a single record, which the distributed ledger achieves. He called proof of work "an arms race" and said that "you can create new cryptographic regimes." He saw decentralized authentication as the real strength of blockchain-style systems.
Later, Ismail said he wanted to "drill into this conflating computational ability with solving mathematics." Cryptography doesn't depend only on solving math, he argued. "You could use AI to also find stronger cryptography. And so I think it becomes an arms race. That arms race goes in both directions."
Wissner-Gross agreed the race runs both ways but said it is "attacker friendly." Establishing a security protocol takes time "in the present way we organize our civilization," he said. Getting everyone to agree on a single cipher suite, the bundle of algorithms that systems use to talk securely, is "a social construction problem, a network effect type problem." He sketched a scenario in which an attacker asks a newly released frontier model to break a cryptosystem and "it can just do it," while "it takes the entire ecosystem a while to upgrade to something new that another AI has recommended." He said that might change in a future of "hot swapping cryptographic protocol suites left and right nonstop."
Nobody resolved it. Ismail said his "inner Alex" was channeling "defensive co-scaling." Dave Blundin, founder of Link Ventures, had to leave for a partner meeting. "You guys can resolve this," he said. "I'll just read your answer when you publish it."