OpenAI’s ETH Claim Does Not Break Bitcoin Cryptography—But the Review Has Just Begun

A dramatic computer-science claim from OpenAI has landed in the crypto world, but Bitcoin holders need to separate the headline from the actual security question.

OpenAI says one of its systems has refuted the Exponential Time Hypothesis, or ETH, a longstanding assumption about how difficult certain logic problems become as they grow.

In this context, ETH has nothing to do with Ether. It is a claim about the worst-case complexity of solving families of Boolean formulas.

The key point for Bitcoin is simple: a faster theoretical algorithm for one broad class of problems is not the same thing as a working attack on Bitcoin mining, private keys, ECDSA, Schnorr signatures, or SHA-256.

What OpenAI is actually claiming

According to CryptoSlate’s detailed review of the announcement, the manuscript claims a deterministic algorithm that counts satisfying assignments for fixed-width Boolean formulas using subexponential time and polynomial space. That would be a major theoretical result if it survives scrutiny, but it would not make every hard problem easy or settle the famous P versus NP question.

The distinction matters because “subexponential” is a mathematical growth rate, not a benchmark showing that today’s computers can recover Bitcoin keys. A practical crypto attack would still require researchers to encode the exact target problem, build a working implementation, measure its cost, and show that it beats existing attack methods at the relevant security parameters.

OpenAI says the work has been checked in Lean, a proof assistant. Even so, outside researchers are not treating that as the final word.

Ryan Williams, Russell Impagliazzo, and Mohan Paturi are organizing an open independent examination of the claim.

An OpenAI model has produced a claimed refutation of ETH. Russell Impagliazzo, Mohan Paturi, and I have been asked to help organize an open and independent examination of this claim. 1/4

— Ryan Williams (@rrwilliams) October 26, 2026

That independent review is not a ceremonial step. The sharpest claim in a theoretical paper can depend on definitions, reductions, hidden constants, or a proof step that looks different when specialists work through it line by line.

The organizers have made clear that their involvement should not be read as certification.

Why Bitcoin’s lock is still intact

Bitcoin mining and Bitcoin spending rely on different concrete tasks. Miners search for a block-header hash below the network target.

Wallets authorize spending with digital signatures tied to secp256k1 keys. Neither task is automatically defeated because a new algorithm improves the worst-case bound for a class of SAT problems.

For the spending side, an attacker would need a practical method to derive a private key from exposed public information or forge a valid signature.

For mining, an attacker would need a measurable advantage in the hash search that produces valid blocks. The OpenAI result, as publicly described, supplies neither demonstration.

Williams also said he specifically asked whether existing cryptosystems were compromised and relayed that they apparently were not. That is encouraging, although it remains a second-hand report of OpenAI’s own check rather than a completed independent cryptographic audit.

In fact, one of the first things I wanted OpenAI to check is that no existing cryptosystems are compromised by the result. Apparently they are safe…

— Ryan Williams (@rrwilliams) October 27, 2026

The real signal for crypto

The right reaction is neither panic nor dismissal. If the result holds, it would show that AI-assisted mathematical research can move the boundary of what experts thought was provably hard.

That is important for every industry built on cryptographic assumptions.

But cryptography does not fall because a neighboring theoretical assumption moves. Researchers still have to connect a general result to a specific construction, demonstrate an attack, and compare its real cost against the security margin.

The number of variables in a SAT encoding is not interchangeable with a 256-bit key. Plugging one familiar number into an unrelated bound does not produce a Bitcoin exploit.

Bitcoin’s security community should watch the independent review closely. A confirmed result could inspire better algorithms, new proof techniques, or practical improvements years from now. It could also turn out to have narrower consequences than the first headlines suggest.

For now, the honest conclusion is straightforward: OpenAI has announced a potentially important complexity-theory result, independent examination is underway, and no public evidence shows that Bitcoin’s signatures or mining system have been broken.

Join the conversation!

We have no tolerance for comments containing violence, racism, profanity, vulgarity, doxing, or discourteous behavior. If a comment is spam, instead of replying to it please click the icon below and to the right of that comment. Thank you for partnering with us to maintain fruitful conversation.