More than one hundred participants and their AI agents spent roughly two months building a quantum circuit for secp256k1, the elliptic curve used by Bitcoin and Ethereum. According to Eigen Labs, the circuit scores more than 50% lower on the Q x T metric than the result Google Quantum AI reported in March 2026. The comparison is numerical: interfaces and accounting conventions differ, so it is not a claim of formal superiority.
The circuit uses 1,151 logical qubits and 1.30 million Toffoli gates, operations among the most expensive on a quantum machine. In March Google disclosed resource thresholds for such a circuit without publishing its construction. Verification was possible through a zero-knowledge proof that confirmed the circuit existed without revealing it.
The starting point was a record-and-replay technique called dialog-GCD, introduced in 2025 by Tanuj Khattar and Craig Gidney. In June André Schrottenloher published an adaptation to elliptic-curve point addition. The verifier released with Google's proof supplied the rest: a program that checks whether a circuit is correct and how many resources it uses, a very clean optimization target for AI agents.
In May Gautham Anant of Eigen Labs built a benchmark that accepts a circuit, checks it, and reports its resource use. Coding agents were connected to the loop: they study the literature, propose a change, compile, evaluate, and repeat. The first internal effort progressed quickly and then plateaued. In early June Eigen Labs opened the repository, evaluator and leaderboard so anyone could bring a different model, prompt or strategy.
According to IEEE Spectrum, the crowdsourced effort matched Google's displayed result within eight hours and surpassed its reported Q x T score in roughly 72 hours; the 72-hour figure comes from that single outlet. Eigen Labs calls the method Open Autoresearch: humans and agents publish evaluator-verified improvements to a public leaderboard.
The case shows how public verification can speed up research once reserved for large labs. For cryptography it also means estimates of the resources needed to break RSA or elliptic curves keep falling, shortening the time available for migration to post-quantum standards. As previously reported, Claude broke HAWK in 60 hours, and the NIST is pushing migration by 2027.
No comments yet. Be the first!