InvestIndependently confirmed2 publishers2 min readPublished
OpenAI's math results leave open whether AI can break the public-key algorithms behind banking
Fortune and a16z crypto both say OpenAI's solutions to hundreds of long-open math problems do not themselves attack any encryption scheme. Fortune still argues the release signals an AI that may find flaws in banking's public-key algorithms; Stanford's Dan Boneh says nobody knows how the models fare at cryptography.
The Investor · Invest desk
Three figures: Boneh says systems used to wait like 10 years before deployment; a derived calculation puts a month-long model run at roughly 120 times shorter than that wait; Dragonfly's Haseeb Qureshi posits a 95% chance that everything is totally fine.
| Figure | Value | Note | Claim |
|---|---|---|---|
| Boneh: past wait to deploy, like | 10 years | 12 | |
| Calc: month run vs wait, roughly | 120 times shorter | 18 | |
| Qureshi's odds all is totally fine | 95 % chance, posited | 5 |
What happened
- Fortune quotes a prominent Johns Hopkins cybersecurity expert as holding that "We might lose public key cryptography."
- Citing the possibility that ECDSA could soon be cracked, one Bitcoin researcher suggested holders "calmly begin planning for 'bunker mode'" by moving funds to secret new addresses.
- Fortune says news of the release created "barely a ripple in the market."
- On the a16z panel, general partner Eddy Lazzarin said, "We didn't learn that elliptic curves are vulnerable in a way that they weren't before."
Why it matters
- exposure Crypto holders have no undo. Fortune says most banks could reverse an AI-powered theft, while blockchain transactions are deemed irreversible, so Bitcoin and Ethereum holders carry a loss a bank would unwind.
- constraint A quiet market is weak evidence of safety. Citing TBPN, Fortune says an attacker with a working break would do better stealing in small amounts than draining a chain, so a break could go unseen.
- decision Chains must choose between moving funds, which other crypto experts call too drastic for now, and Qureshi's route of putting speed ahead of decentralized purity and mandating a signature system that keeps a copy of transactions.
- precedent If Boneh is right, a new assumption can be stress-tested by a model in about a month where deployment used to wait about 10 years, a 120-to-1 compression, and failed attacks become the evidence.
Fortune and the cryptographers are answering different questions. Fortune's is what the release signals: it describes documents that solve hundreds of problems top mathematicians had struggled with for decades [1] as the arrival of "a new species of mathematical super-genius" that "may be capable of finding flaws" in the algorithms behind banking passwords and email accounts [20]. The a16z panel's is what the release contains. Dan Boneh, who heads the applied cryptography group at Stanford, said "there are no questions here on cryptography" among the results [8]. Asked whether that means the models are "really bad at cryptography", he said: "We don't know." [7]
The cryptographers' case for a low ceiling is a belief about the world, and they say so. Justin Thaler, a research partner at a16z crypto, said researchers generally believe no efficient algorithm exists for these problems, and that "if there is not one, it doesn't matter how smart the AI is. It cannot find the nonexistent algorithm." [10] Boneh was explicit that the belief is unproven. Proving the problems hard is tied to the P versus NP question, and "None of the problems that were posted this week come even close to resolving the P versus NP problem." [11]
That leaves three outcomes. No efficient algorithm exists and the models find nothing; one exists and a model finds it; or one exists and the models are not yet strong enough. Only one source gives a probability. Dragonfly's Haseeb Qureshi, in a post titled "Against Cryptographic Doomerism", puts 95% on "everything is totally fine" [5]. The remainder is 5% [19].
Fortune rules out one route to harm: it says the release will not hasten Q-day, when quantum computers crack popular encryption [2]. The quantum migration work helps in part. Eddy Lazzarin, a general partner at a16z crypto, said, "So we have already actually engaged in and prepared and deployed many of the mitigations you'd need for those exact concerns for a different threat." [14] The a16z write-up adds that the overlap has limits [15].
All five people listed on the a16z episode hold a role at a16z crypto, Boneh as a special research advisor [16].
In our view the release makes the question cheaper to test and leaves the answer where it was, and no source here measured whether the odds moved. Evidence against us would be a published AI-found attack on an elliptic-curve scheme. Evidence for us would be a long run of such attempts that fail.
What to watch
- A published AI-found weakness in an elliptic-curve or other deployed public-key scheme, which would settle the question against the cryptographers.
- Any blockchain that adopts a Qureshi-style signature mandate or governance change.
- Whether any lab publishes cryptography problems among its math results, since Boneh said these contained none.
Clarity's read
What the record supports and how the coverage leans. The claims behind it follow.
Reality
- Evidence38
- Adoption
- Insufficient
- Hype gap+35
- Incentives55
- Confidence50
Claim ledger
Ranked by verification strength, evidence, and original report placement.
- [1]
Fortune says the OpenAI documents reveal the solution to hundreds of math problems that top mathematicians have struggled for decades to solve.
- [2]
Fortune says the OpenAI documents do not describe a new way to hack global computers and will not accelerate Q-day, when quantum computing cracks popular encryption algorithms.
- [3]
Citing the possibility that ECDSA could soon be cracked, one Bitcoin researcher suggested crypto holders 'calmly begin planning for bunker mode' by moving their funds to secret new addresses.
- [4]
Fortune says most banks and financial institutions would be in position to reverse transactions carried out by an AI-powered hacker, but blockchain transactions are deemed irreversible, making Bitcoin and Ethereum likely prime targets.
- [5]
Dragonfly's Haseeb Qureshi, in a blog post titled 'Against Cryptographic Doomerism', posits a 95% chance that 'everything is totally fine'.
- [6]
Qureshi says blockchains must improve governance and for now prioritize speed over decentralized purity, and calls for updates mandating a signature system that leaves a copy of everyone's transactions so a chain can survive a wipeout.
- [7]
Asked whether the lack of cryptography questions in the results means the models are really bad at cryptography, Dan Boneh said: 'We don't know.'
- [8]
Dan Boneh, a Stanford professor who heads the applied cryptography group, said of the recently published results: 'there are no questions here on cryptography'.
- [9]
Eddy Lazzarin, general partner at a16z crypto, said: 'We didn't learn that elliptic curves are vulnerable in a way that they weren't before.'
- [10]
Justin Thaler, a research partner at a16z crypto, said researchers generally believe there is no efficient algorithm for these problems, and that 'if there is not one, it doesn't matter how smart the AI is. It cannot find the nonexistent algorithm.'
- [11]
Boneh said cryptography must assume certain problems are hard because proving it is tied to the P versus NP problem, and that 'None of the problems that were posted this week come even close to resolving the P versus NP problem.'
- [12]
Boneh said that in the past, when someone made an assumption and built a system on it, 'we'd have to wait like 10 years' before the system could be deployed.
- [13]
Boneh said that now 'we can just let a model run for a month, try to break it, and if the model can't break it, boy, do we have a lot of confidence that the assumption is valid.'
- [14]
Eddy Lazzarin said: 'So we have already actually engaged in and prepared and deployed many of the mitigations you'd need for those exact concerns for a different threat.'
- [15]
The a16z write-up says the overlap between quantum-driven migration of some systems and AI risk has limits.
- [16]
The five people listed on the a16z episode, Boneh (special research advisor), Lazzarin (general partner), Thaler (research partner), Liz Harkavy (investment partner) and Sonal Chokshi (editor in chief), each hold a role at a16z crypto.
- [17]
Other crypto experts have replied that moving funds to new addresses is too drastic for now.
- [18]
A 10-year wait is about 120 months, so a month-long model run is roughly 120 times shorter than the wait Boneh described.
- [19]
Qureshi's 95% chance that everything is fine leaves a 5% remainder for every other outcome.
- [20]
Fortune says the documents signify the arrival of 'a new species of mathematical super-genius' that 'may be capable of finding flaws' in the algorithms used to encrypt banking passwords, email accounts and the rest of digital life; those algorithms have long been regarded as sound by top cryptographers.
- [21]
Fortune quotes a prominent cybersecurity expert at Johns Hopkins as holding: 'We might lose public key cryptography.'
ReportedContestedSource: Fortune, quoting a Johns Hopkins cybersecurity expert2 sources— create a free account to open themView cited source - [22]
Fortune says news of the 'Mathpocalypse' created barely a ripple in the market.
- [23]
As TBPN noted, an attacker holding a working break would be foolish to carry out a massive heist that would destroy the value of the crypto; the better play is a series of smaller, subtler thefts, so a flaw may not be discovered right away.
Sources
2 independent publishers whose own reporting we read for this story.
- fortune.comA ‘Mathpocalypse’ is freaking out cyber and finance experts. Should you be worried?
1 article · October 12, 2026
- open.substack.comCan AI really break cryptography? Experts weigh in
1 article · October 10, 2026
Topics and entities
Follow any of these and your For You feed starts watching them — no settings page required.