Theoretical cost of a quantum attack on cryptography drops by 86%

The theoretical cost of a quantum attack has just dropped. More than a hundred participants, supported by AI agents, reduced by 86.1% the resources required for one of the calculations that would one day make it possible to attack the signatures used by Bitcoin and Ethereum. No private keys have been recovered and no current quantum computer can steal cryptocurrencies thanks to this work. The result nevertheless shows that improving programs can bring the threat closer, even without major improvements in hardware.

The ECDSA.Fail competition, launched by Eigen Labs, aimed to improve a calculation used numerous times in Shor's algorithm. This could allow a sufficiently powerful quantum computer to find a private key from a public key, then sign transactions in place of its owner.

Bitcoin and Ethereum notably use the secp256k1 cryptographic system to authorize transactions. The competition, however, did not seek to break a complete address. The participants only worked on one of the calculations necessary for this attack.

To compare their proposals, the organizers created a score combining the size of the machine needed and the amount of work it should do. The lower this score, the less resources the circuit theoretically requires.

According to the report published on arXiv, the score increased from 10.75 billion to 1.496 billion between the launch of the competition and the data cut-off on July 26. The best circuit uses 1,151 logical qubits, the computing units of an error-corrected quantum computer, and about 1.3 million complex operations.

The result is more than 50% below the score of 2.993 billion published by Google Quantum AI in March. The authors emphasize, however, that the two models do not follow exactly the same calculation rules. The comparison therefore gives an order of magnitude, but is not enough to prove that one method formally outperforms the other.

After July 26, a new proposal lowered the score further to 1.259 billion. Another circuit reduced the minimum size to 813 logical qubits, at the cost of a much larger number of operations.

More than 400 improvements were validated during the competition. The AI ​​agents were mainly used to modify the code, run the tests and quickly explore many small optimizations. Humans have been more likely to choose research avenues and design important changes.

The report therefore does not allow the 86.1% drop to be attributed to AI alone. Rather, it shows how researchers and autonomous software can move forward together when a result can be verified automatically.

This work remains very far from a real attack. They do not integrate the entire Shor algorithm nor all the constraints allowing the errors of a quantum machine to be corrected. The number of physical components required would also depend on the type of computer used: it cannot be deduced directly from the 1,151 logical qubits announced.

The preparation has nevertheless begun. NIST, the American institute responsible for technological standards, has already selected several protections designed to resist quantum computers. Its draft calendar recommends gradually abandoning certain traditional systems after 2030, then no longer using them after 2035.

Galaxy has also committed up to $5 million to an initiative dedicated to the quantum preparation of Bitcoin. Another consortium bringing together BlackRock, Coinbase and Strategy has promised $15 million over three years to finance the security and development of the network more broadly.