🔍 Read the full analysis: AI Mathematics And Quantum Computing Raise New Encryption Questions on ThorstenMeyerAI.com
Get business pricing on tech for your team
- Business-only prices and quantity discounts
- Tax-exempt purchasing
- Multiple users, one account, clear invoices
TL;DR
OpenAI reported producing 722 mathematical manuscripts with an internal AI model, while researchers in cryptography and cryptocurrency warned that AI-assisted discoveries could challenge assumptions about computational hardness. No encryption scheme has been shown to be broken, and experts differ on how soon any practical risk could emerge.
OpenAI reported on 6 October that an unreleased internal AI model had produced 722 mathematical manuscripts across 372 families, prompting renewed concern that AI could help find algorithms that challenge cryptographic assumptions. Researchers have not reported a working attack on encryption, but Ethereum researcher Justin Drake and co-founder Vitalik Buterin have raised different concerns about the risks to current and proposed systems.
OpenAI said the manuscripts came from work on roughly 4,000 mathematical problems, with an average of about three hours of ChatGPT Pro compute per result. The claims include results involving the Unique Games Conjecture, Hilbert’s tenth problem over the rationals and the Riemann zeta function. These are reported claims, not a set of independently settled results: checking the work is ongoing, and OpenAI withdrew a claimed proof concerning the Hodge conjecture for products of K3 surfaces after a reported sign error.
For cryptography, the concern is not that these manuscripts have broken encryption. It is that some recent results challenge assumptions about how efficiently certain problems can be solved. Computer scientist Scott Aaronson highlighted claims involving faster integer multiplication and Fourier transforms. Separately, a paper by Virginia Vassilevska Williams and Josh Alman described a new approach to 3SUM, with a key idea attributed to an Anthropic model. These algorithmic developments do not by themselves establish an attack on a cryptographic protocol.
Aaronson also noted that cryptographic results were absent from OpenAI’s published collection. He said his sources indicated AI companies were discreetly testing whether internal models could break important protocols. That account has not established that any such model has succeeded. Public comments from Drake and Buterin have focused on possible future weaknesses, rather than announcing a verified compromise.
The old map is gone: AI mathematics, quantum computers and the cryptography holding up finance and defence
For a decade the plan was simple: elliptic curves doomed by quantum; lattices safe; hashes safe. Nothing has been broken. But a second threat has arrived that doesn’t respect those borders — AI producing new mathematics faster than any human community, against assumptions that are believed, not proven.
Now: on borrowed time — possibly shorter than the quantum countdown suggests.
Now: unproven against AI — and the destination most of the world is migrating to.
Now: reminded estimates move — BSI advised against new deployments on 1 Oct 2026.
Now: safest ground available — not a guarantee.
~n log0.9999999999999 n — a barrier many thought fundamental (OpenAI, claimed)
Overturns a half-century conjecture. Williams & Alman; key idea from an Anthropic model
“Conspicuous by its absence” (Aaronson) — labs reportedly testing crypto “gingerly and discreetly”
ECDSA could break before Q-day, “in the worst case in months not years.” Move funds to never-signed addresses. ~6M BTC sit behind exposed keys.
The new risk is the destination of the migration. Hash-only where possible; “much more paranoid” lattice params; ×10 key sizes long-term. Doesn’t recommend anyone scramble.
“No evidence whatsoever” that elliptic-curve assumptions are close to failing.
Classical breaks could reach “quantum-safe” schemes — but don’t treat a two-year scenario as a date.
Known to IBM and the NSA designing DES (~1974); public via Biham & Shamir (~1990); confirmed by Coppersmith (1994).
Invented at GCHQ — RSA- and Diffie–Hellman-equivalents — and kept secret for over two decades.
No crypto in 722 manuscripts. Found and withheld? Not posed? Posed and failed? Indistinguishable from outside.
Traffic recorded today is decrypted when a break arrives. For secrets that must last 25+ years, a break in 2035 is a break today. A state that finds one won’t announce it — it will mine its archives.
Signatures can be built from hashes. Encryption and key exchange need a trapdoor with structure — lattices, codes or group theory. Defence can only choose which structure, how much margin, how many combined.
Every date was set against quantum hardware forecasts with visible warning. The AI threat offers none.
“ML-KEM everywhere” means starting over if lattices weaken. “We can swap algorithms” doesn’t.
Blockchains show a classical break first — exposed keys and balances are public. Monitor dormant exposed addresses.
Every algorithm, key, certificate, protocol.
PQ + classical, as BSI requires.
Firmware, updates, long-term keys.
Highest sets; evaluate FrodoKEM.
More than one mathematical family; HQC coming.
Swap algorithms without rebuilding.
Forward secrecy, rotation, hidden keys.
Buterin: lost more in botched migrations than in all hacks.
Nothing has been broken, and the sceptics are right that there’s no evidence elliptic curves or lattices are about to fall. But the map has changed: elliptic curves on borrowed time, lattices unproven against AI, codes reminded that estimates move, hashes the safest ground available. For finance, intelligence and defence the answer is the same whichever threat arrives first.The quantum threat comes with a countdown. The AI threat may arrive as a silence — an empty folder where a paper should have been. The winners will be those who can change their algorithms fastest.
Why Hardness Assumptions Matter
Encryption depends on mathematical problems believed to be difficult to solve with available methods. Those beliefs are supported by years of analysis and testing, but they are not general proofs that no faster algorithm exists. If an unexpected method made a problem tractable, systems built on that problem could require replacement, regardless of whether the discovery came from a person or an AI model.
The practical stakes reach beyond cryptocurrency. Public-key cryptography helps protect financial transactions, communications and government systems. A discovery could matter most if it were usable but kept secret: defenders might not know that an assumption had failed until information or assets were exposed. That possibility is a risk scenario, not evidence that a hidden break currently exists.
AI could also help security researchers test systems and identify weaknesses earlier. The same capacity that raises concern may support defensive review. The central issue is whether models can produce reproducible, efficient algorithms with consequences for real protocols—not simply whether they can generate sophisticated-looking proofs.
quantum computing encryption tools
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Quantum Migration Meets AI Research
The current migration debate began with a different threat. A sufficiently capable, fault-tolerant quantum computer running Shor’s algorithm could undermine RSA and elliptic-curve cryptography. Governments and companies have therefore been preparing to move to post-quantum systems. In August 2024, the U.S. National Institute of Standards and Technology standardized ML-KEM for establishing encryption keys, ML-DSA for digital signatures and the hash-based signature standard SLH-DSA.
Quantum risk has a hardware component that can be monitored, even though the arrival date for a machine capable of breaking widely used cryptography remains uncertain. The AI concern described by the sources is different: a better algorithm might run on conventional computers, and its discovery could remain private. That makes the threat harder to assess from public information. It also does not follow that lattice-based standards are broken; their security is based on different mathematical assumptions from RSA and elliptic curves.
Cryptocurrency has made the debate visible because some blockchain addresses expose public keys after use. Drake urged the industry to calmly plan for a possible “bunker mode,” including moving funds to addresses whose public keys have not been exposed. Reports cited in the source material put about six million bitcoin in addresses with exposed public keys. That figure describes exposure, not an estimate of funds already recoverable by an attacker.
““IMO it is now reasonable to brace for the possibility that ECDSA breaks before qday, in the worst case in months not years.””
— Justin Drake, Ethereum Foundation researcher
No Verified Encryption Break
No cryptographic protocol has been shown to be broken by the mathematical work described here. The status of the manuscripts varies, and verification is still needed; the withdrawn Hodge-conjecture claim illustrates why publication or model output should not be treated as proof without scrutiny.
It is also unclear whether AI systems can find a practical algorithm that weakens RSA, elliptic-curve cryptography or lattice-based standards, and whether any such discovery would be disclosed. The source material describes discreet testing based on information from Aaronson’s sources, but provides no public demonstration or independent confirmation of a successful cryptographic attack. The timing and scale of any real-world risk remain unknown.
Verification and Migration Decisions
Mathematicians and computer scientists will need to check the reported proofs and algorithmic results, including whether they can be independently reproduced and what their speedups mean in practice. AI companies’ testing claims will require public evidence before readers can judge whether the work has produced a cryptographic capability rather than a warning about one.
Meanwhile, organizations can continue planning migration to standardized post-quantum cryptography without assuming that the new standards are either infallible or already compromised. Cryptocurrency users should distinguish Drake’s call for contingency planning from Buterin’s advice not to rush into wallet changes. The next meaningful update would be a verified result showing a reproducible attack, or a detailed evaluation from researchers demonstrating why a particular cryptographic assumption needs to be revised.
Key Questions
Has AI broken encryption?
No verified encryption break has been reported in the source material. The concern is that AI could help discover algorithms that challenge mathematical assumptions behind cryptography.
What did OpenAI report?
OpenAI reported that an unreleased internal model produced 722 mathematical manuscripts across 372 families, drawing on work involving roughly 4,000 problems. The claims require checking, and at least one claimed proof was withdrawn.
How is the AI concern different from the quantum threat?
A sufficiently capable quantum computer could use Shor’s algorithm against RSA and elliptic-curve systems. The AI concern is about finding better algorithms that might run on ordinary computers, with no hardware countdown that observers can easily track.
Are post-quantum standards known to be unsafe?
No. The concerns raised in the source are not evidence that NIST’s lattice-based standards, including ML-KEM and ML-DSA, have been broken. Their security remains subject to ongoing mathematical analysis.
Should cryptocurrency users move their funds now?
Drake urged planning for a possible protective response, while Buterin said he did not recommend scrambling to move funds immediately. The source does not establish a current attack or a general instruction for users to move assets.
Source: ThorstenMeyerAI.com
Halloween Picks
halloween
As an affiliate, we earn on qualifying purchases.
