Morning Minute: AI Agents Cut BTC Quantum Attack Benchmark by 86%

More than 100 researchers working with AI coding agents reduced a benchmark for a core step in a quantum attack on Bitcoin and Ethereum from 10.75 billion to 1.496 billion between late May and July 26, a decline of roughly 86%. The result, published this week, comes from a competition-style effort called ECDSA.Fail and was presented by a consortium of crypto security teams in a paper using a method they name Open Autoresearch.

By AI NewsroomPublished 17 minutes agoUpdated 17 minutes ago0 views

Morning Minute: AI Agents Cut BTC Quantum Attack Benchmark by 86%

Why It Matters

The finding shows a major attack-related resource metric falling rapidly over a two-month period while industry efforts to harden networks follow fixed multi-year timetables (for example, the Ethereum Foundation’s December 2029 target and NIST’s draft deprecation dates). That gap in tempo could affect how protocols and firms prioritize quantum-resistant upgrades and funding.

Key Facts

Benchmark reduction: From 10.75 billion to 1.496 billion (≈86% decrease)
Timeframe of improvement: Late May to July 26
Paper published: This week (paper authored by multiple crypto security teams)
Competition name: ECDSA.Fail (run by Eigen Labs)
Scoring method: Circuit designs scored by logical qubits × Toffoli gates (Toffoli is an expensive quantum operation)

A coordinated effort of more than 100 researchers working with AI coding agents dramatically lowered a benchmark tied to a key step in a potential quantum attack on Bitcoin and Ethereum. The metric tracked by the ECDSA.Fail competition fell from 10.75 billion to 1.496 billion between late May and July 26, representing roughly an 86% reduction in the measured resource score.

ECDSA.Fail, organized by Eigen Labs, evaluates circuit designs by multiplying the number of logical qubits by the count of Toffoli gates, a costly quantum operation. Early leading submissions used about 1,151 logical qubits with roughly 1.3 million Toffoli gates; later entries managed to push gate counts below one million. The paper’s top score was about half of a March benchmark published by Google Quantum AI, although the two results use different counting approaches and aren’t directly comparable.

The research was published by teams from Eigen Labs, Trail of Bits, StarkWare, Theta Labs, MultiVM Labs, and the Ethereum Foundation. The authors describe their process as Open Autoresearch: humans and AI agents iterating on a shared, measurable target with an automated verifier in the loop. The groups say the approach lets many contributors rapidly search for improvements against the same objective.

The work arrives amid a broader industry push to prepare for quantum threats. The Ethereum Foundation has set a December 2029 deadline to make transactions, validators, and storage quantum-resistant, and StarkWare recently published what it calls the first quantum-safe Bitcoin transaction on mainnet. Other projects are proposing protocol changes, firms like Galaxy announced up to $5 million in commitments, and a coalition of nine firms including BlackRock and Coinbase pledged $15 million over three years. Meanwhile, NIST’s draft guidance proposes deprecating classical public-key algorithms at the 112-bit level after 2030 and disallowing them after 2035.

The authors and observers emphasize this is not an immediate emergency: no funds are at risk today and the paper does not predict an imminent “Q-Day.” Still, the results highlight that attack-research progress can be accelerated when many people deploy AI agents against a measurable target, while many defensive programs operate on fixed multi-year timetables. That dynamic may influence how quickly protocols and organizations accelerate quantum-hardening plans.