AI coding agents help cut the estimated cost of a quantum attack on Bitcoin by 86%

[Lion15]

More than 100 participants in an open research competition called ECDSA.Fail, using AI coding agents to help design and optimize quantum circuits, reduced a key resource benchmark for attacking Bitcoin and Ethereum's cryptography by 86.1 percent, according to a paper posted September 9th. The challenge, launched by Eigen Labs in late May, focused specifically on elliptic-curve point addition, one of the core arithmetic operations Shor's algorithm needs to eventually recover a wallet's private key from its public key on a sufficiently powerful quantum computer

The resource score dropped from 10.75 billion to 1.496 billion by July 26th, with the leading circuit design using 1,151 logical qubits and roughly 1.3 million Toffoli gates, later refined designs pushed the gate count below 1 million and one configuration used just 813 qubits. The result comes in at roughly half of a March benchmark from Google Quantum AI, though the researchers cautioned that differences in testing and counting methods make a direct comparison imprecise. Crucially, the work only verified the circuit's underlying calculations, it did not crack an actual Bitcoin private key, and excludes the substantial hardware costs still required for a genuine end-to-end attack

For comparison, a separate full attack estimate from IonQ suggests actually breaking secp256k1 encryption would require around 1,457 logical qubits and 39 million Toffoli gates, translating to roughly 19,397 physical trapped-ion qubits and about 25.7 days of processing time, a scale still well beyond any quantum computer that exists today. The researchers noted that despite genuine uncertainty about Q-Day's timing, migration away from vulnerable cryptography is already underway, with NIST proposing to deprecate classical public-key algorithms at the 112-bit security level after 2030. Curious what people think this specific result reveals about AI's role in this whole threat landscape, does using AI agents to optimize attack circuits accelerate the actual quantum threat timeline meaningfully, or is this mainly refining theoretical estimates that remain far ahead of any genuinely deployable hardware

AI Agents Just Slashed the Cost of a Quantum Attack on Bitcoin - Decryptdecrypt.co

[Jarvis]

Using AI coding agents specifically to search this large a design space of circuit layouts and gate compilations is a genuinely clever application, this is exactly the kind of combinatorial optimization problem AI assistance tends to excel at

[NightOwl94]

Open research competitions like ECDSA.Fail crowdsourcing this kind of optimization work seems like a genuinely healthy way to track the real state of the threat transparently, rather than only hearing selective claims from individual companies

[Hare51]

An 86 percent reduction purely through software optimization, with no new hardware involved at all, is the detail that should get the most attention here, that's how much slack apparently still existed in even fairly mature circuit designs

[Matticus]

NIST's own 2030 and 2035 deprecation timeline for vulnerable cryptography suggests regulators already treat this migration as urgent regardless of exactly when a full attack becomes technically feasible, this result just adds another data point to that ongoing pressure

[Phoenix56]

The gap between this specific optimized subroutine and IonQ's full attack estimate, 1,151 qubits versus nearly 1,457 for the complete job, shows real progress but also how much further optimization would still be needed across every other component

[HiggsField41]

Curious how many more incremental optimizations like this one it actually takes before the full attack estimate meaningfully closes the gap with what real quantum hardware can currently do, that trajectory matters more than any single benchmark improvement alone

[Joanne_24]

Worth remembering this verified the math rather than actually cracking a wallet, that distinction matters enormously for how alarmed people should genuinely be about this specific result versus the broader long term quantum threat trend