OpenAI Says Its AI Refuted the Exponential Time Hypothesis, and Researchers Say Bitcoin's Signatures Are Not Affected

OpenAI said on October 26 that one of its AI systems had disproved the Exponential Time Hypothesis, the long-standing assumption in complexity theory that problems such as 3-SAT cannot be solved meaningfully faster than brute force in the worst case. The paper describes a deterministic algorithm that counts 3-SAT solutions in subexponential time, according to KuCoin News. OpenAI first posted it on October 17 and revised it on October 25. Researchers have organized an independent review, and the result is not yet verified.

The crypto market's first reaction confused two different meanings of "ETH". CryptoSlate points out that ETH here refers to the hypothesis, not Ethereum. It also notes that a theoretical speedup for satisfiability problems does not by itself produce a practical attack on cryptography. The elliptic-curve signatures that protect Bitcoin and Ethereum, ECDSA and Schnorr, rest on the hardness of the discrete-logarithm problem, which a faster 3-SAT counting algorithm does not break. Researchers cited in coverage of the paper said practical attacks on on-chain keys remain unlikely without further, separate advances.

On the defensive side, the evidence is specific. Aztec, the privacy-focused Ethereum layer-2 (L2) network, has disclosed that its core contributors found a critical flaw in the Alpha V5 proving system on July 27 through internal AI-assisted auditing. The flaw could let an attacker build a proof that passes verification for a transaction the network should reject, allowing a state change outside the rules. Aztec told users to treat funds and applications on Alpha V5 as exposed until incident response ends, and it plans the fix for V6. Validators that re-execute transactions would not catch a flaw of this kind, because they trust the proof rather than checking it independently. Aztec has separately raised its total bug bounty to $2 million.