AI Agents Just Slashed the Cost of a Quantum Attack on Bitcoin

An open competition called ECDSA.Fail brought together more than 100 contributors using AI coding agents to optimize quantum circuits for secp256k1, the elliptic curve used in Bitcoin and Ethereum signatures. Participants focused on the computational subroutine needed to recover a private key, and documented designs and tests in a paper released by the organizers and collaborators. The contest scored designs by multiplying a circuit’s logical qubits by its number of Toffoli gates, a costly quantum operation; that product served as a combined resource benchmark. Over the course of the challenge the best score fell sharply—from 10.75 billion down to 1.496 billion by July 26—representing an 86% reduction in the measured resource requirement. The top circuit reported used 1,151 logical qubits and around 1.3 million Toffoli gates, and a later design reduced the gate count to below one million. The authors emphasize important caveats: the experiments verified that the circuits implement the targeted calculation but did not recover any real private keys, and the reported scores exclude the hardware and other costs of mounting a complete, end-to-end quantum attack. The paper also cautions against direct head-to-head comparisons with prior benchmarks—its score is about half of Google Quantum AI’s March figure, but differences in counting and testing methods limit the validity of a direct comparison. Beyond the technical results, the project served as a testbed for a collaborative workflow in which humans and AI agents iteratively propose, implement and validate improvements under a shared objective. The report appears amid broader industry and standards activity: NIST has advanced post-quantum replacements and proposed timelines to deprecate certain classical public-key algorithms after 2030 and disallow them after 2035, while firms including Galaxy Digital and a consortium of nine companies have pledged millions to fund quantum-resilience research.