Google Quantum AI reduced a critical class of quantum errors roughly fortyfold on its Willow chip, according to a paper published this week in Science. The technique targets magic-state distillation, which by some estimates consumes up to 95 percent of a quantum machine's operational capacity. It produced outputs at 99.99 percent fidelity after filtering out 92 percent of attempts that failed, sharpening — but not resolving — a cryptographic question the digital-asset industry had treated as a distant threat.

How far is cryptography from being cracked?

Breaking an RSA-2048 key requires error rates of roughly one in ten billion — about a million times cleaner than Google's result. The qubit estimates have compressed sharply: in 2019, cracking RSA-2048 was estimated to need 20 million noisy qubits; by 2025 that figure had fallen below one million. A March 2026 Google whitepaper put the threshold for breaking Bitcoin's elliptic-curve cryptography at under 500,000 physical qubits. Willow has 105.

The takeaway: the gap remains measured in millions of qubits and orders of magnitude in fidelity, not in months.

What was Justin Drake's actual warning?

Eight days before the Google result, Ethereum researcher Justin Drake urged large holders to move coins to fresh addresses that had never signed a transaction. The risk he flagged was not quantum at all. Drake has argued that AI-assisted mathematics will break elliptic-curve cryptography on ordinary hardware "in months not years," and that a signed transaction exposes a public key, leaving older addresses comparatively exposed. The Google paper, in his framing, made him look less paranoid and more early.

The takeaway: signing, not holding, is the hazard; the threat model is mathematical software, not Willow-class silicon.

Who is preparing, and on what timeline?

Coinbase spent the past week arguing that cryptographic migration has to begin now. Sui is already running quantum-safe transactions on a test network. Ethereum, Zcash and Near have published quantum-resistance roadmaps. None has named a hard deadline for retiring current signature schemes, and a network-wide upgrade on Bitcoin alone would carry substantial coordination costs across miners, custodians and wallet vendors.

The takeaway: the industry has moved from whitepaper to testnet, but no major chain has committed to a switchover date.

What does the trajectory actually suggest?

Google's achievement is a fortyfold reduction in one error class, not a six-order-of-magnitude leap in machine capability. The lesson of the past seven years is that cryptographic thresholds have not been stationary. Each published estimate has moved the goalposts closer, with the noisy-qubit requirement for breaking RSA-2048 falling from 20 million in 2019 to under one million by 2025. Custodians are budgeting accordingly, and the de facto industry deadline will likely be set by whichever major institution moves first under regulatory or fiduciary pressure rather than by any protocol's stated roadmap.

Until either a quantum machine reaches the qubit counts Google's own 2026 paper names, or AI tools demonstrably factor a production elliptic-curve key, the migration remains precautionary; Coinbase's public posture this week suggests the industry's response window has begun to shrink.