Google's Quantum Breakthrough Cuts Magic State Errors 40-Fold

Google showed on real hardware the kind of cheap magic state production that its own estimate for breaking bitcoin's signatures already assumes. No new estimate came with it.

NewsBlock · 971,7278 min read
AI-assisted

AI helped with research and writing. The editor chose topic and angle, decided what stays in and approved the article. Sources at the end.

How we work

Google's latest quantum breakthrough, published in Science on October 15, cut the error rate of so-called magic states by a factor of about 40 on the company's Willow processor. For bitcoin, the result came without a new threat estimate: Google's own March calculation for breaking bitcoin's signatures already budgets for cheap magic states, and no machine able to run that attack exists.

What Google demonstrated

A team led by Emma Rosenfeld and Craig Gidney of Google Quantum AI ran a technique called magic state cultivation on Willow, a superconducting chip. According to the paper, cultivation reduced the error of the prepared state about 40-fold compared with states that skipped the process, reaching a logical fidelity of 99.99 percent.

Cultivation works like a filter. The chip injects a noisy state into a small patch of qubits, checks it with a series of measurements and discards every attempt that shows an error. In the experiment, 8 percent of attempts survived. The team also moved surviving states from a distance-3 color code into a distance-5 surface code, the error-correcting layout Google plans to scale, and showed that they kept their quality in the transfer.

Google calls the work the first demonstration, on real hardware, of a practical solution to one of quantum computing's most severe resource bottlenecks. The company also says cultivation still has to be shown at a larger scale and that plenty of technical challenges remain. The experiment first appeared as a preprint in December 2025.

Why magic states matter

An error-corrected quantum computer handles most operations cheaply. A few operations, known as T gates, need a special resource called a magic state, and each one is used up when the gate runs. Making these states reliably has been one of the most expensive parts of a fault-tolerant machine for two decades. Google's announcement cites estimates in which up to 95 percent of a quantum computer's capacity would go into producing them.

Gidney, Noah Shutty and Cody Jones proposed cultivation on paper in September 2024. Their simulations put its cost at roughly that of a single error-corrected CNOT gate of the same reliability and suggested that the older method, magic state distillation, may never be needed in practice. The new experiment tests that proposal on a chip.

What the result means for bitcoin

Bitcoin authorizes every spend with a digital signature, using ECDSA or Schnorr signatures on the secp256k1 curve. A large, error-corrected quantum computer running Shor's algorithm could derive a private key from its public key. That is the real quantum risk to bitcoin, and it concerns signatures rather than mining. Running Shor's algorithm on bitcoin's curve takes tens of millions of Toffoli gates, and each of them consumes magic states.

In March, Ryan Babbush, Craig Gidney and seven co-authors estimated that the attack needs fewer than 1,200 logical qubits and 90 million Toffoli gates, or fewer than 1,450 logical qubits and 70 million Toffoli gates. On a superconducting machine with an error rate of one in a thousand, that would take fewer than half a million physical qubits and run in minutes. A machine that precomputes half of the work could recover a key about 9 to 12 minutes after the public key becomes visible.

In our reading of that paper, the cost of magic states is already accounted for. The authors assume that one T state can be produced in 50,000 qubit-rounds and set aside 25,000 physical qubits for the job, about 5 percent of the half million. Gidney's 2025 estimate for breaking RSA-2048 with fewer than a million qubits names cultivation among the techniques behind its reduction. The new experiment checks on hardware the kind of cheap magic state production that these estimates already assume. It does not lower them, and Google published no new figure for bitcoin alongside it. A 40-fold lower error rate on one prepared state also does not mean an attack needs 40 times fewer qubits.

Where a public key is visible decides when a coin is exposed. Most bitcoin addresses show only a hash of the public key until the coins are spent, as our guide to bitcoin addresses explains. Google's March paper counted about 6.9 million bitcoin whose keys are already exposed through old address types, Taproot outputs or address reuse. Glassnode arrived at 6.04 million in May, and the gap comes from method and date, not from a disagreement about the risk. All other coins are exposed only while the transaction spending them waits in the mempool, which is why a key recovery time close to bitcoin's average 10-minute block interval matters. Our analysis of how the quantum threat to bitcoin is measured follows these estimates over time.

Research on the defense side

This week, Blockstream researcher Jonas Nick drew attention to Compact Shielded CSV, a preprint by researchers at Imperial College London and the company Orbs that was posted to arXiv in September. It describes private payments designed for a world with quantum computers. Post-quantum signatures are much larger than the ones bitcoin uses today, so the design keeps them off the chain. Users pass the proofs to one another under a model called client-side validation, and the chain records only commitments. For a transaction with one input, that comes to three hashes of 32 bytes each, 96 bytes in total.

The paper describes a separate blockchain, not a change to bitcoin. It builds on Shielded CSV, which Jonas Nick, Liam Eagen and Robin Linus proposed in 2025 as a privacy layer on top of bitcoin, and whose Schnorr signatures are not post-quantum. The new preprint does not specify a consensus mechanism or how a full implementation would handle reorganizations of the chain.

Big news, but a quantum computer that can break bitcoin is still far away

Google's result is a real step. A technique proposed on paper in 2024 now works on a chip, and it attacks one of the costs that make large quantum computers so hard to build. The distance to an attack on bitcoin remains large all the same. Willow prepared individual magic states in codes of distance 3 and 5. The attack Google estimated in March needs 1,200 to 1,450 error-corrected logical qubits running tens of millions of gates in sequence. In April, Project Eleven awarded its Q-Day Prize for breaking a 15-bit elliptic curve key with a variant of Shor's algorithm on quantum hardware. Bitcoin's keys have 256 bits. Google itself says cultivation still has to be demonstrated at scale, and nothing in the paper describes a machine that could derive a bitcoin key.

Work on the defense has started as well. The draft BIP-360 would add an output type, Pay-to-Merkle-Root, which removes the Taproot spending path that leaves a public key on chain. Any such change needs broad agreement among the people who run bitcoin's software, and that makes migration slow. Whether swapping the signature scheme would change bitcoin itself is the question of our essay on what a new signature would and would not alter. The authors of Google's March paper urge every affected network to begin the move to post-quantum cryptography without delay.

There is also a dissenting view from physics. Oxford physicist Tim Palmer argues in a paper published in PNAS in March, and in a companion preprint, that the advantage of quantum algorithms built on the quantum Fourier transform, Shor's among them, ends beyond roughly 1,000 logical qubits. His work deals with RSA-2048 and does not mention bitcoin. Applying it to bitcoin's curve is a reading, not his claim, and his theory departs from standard quantum mechanics. Palmer says it can be tested within five years.

Frequently Asked Questions

No. Willow prepared individual magic states in small error-correcting codes. Google's own estimate for breaking bitcoin's signatures calls for 1,200 to 1,450 logical qubits running tens of millions of gates, and no machine of that size exists.

No. The figure describes the quality of one prepared magic state. Google's March estimate for bitcoin already assumed cheap magic states, and the company published no new resource estimate for bitcoin with this result.

Coins whose public key is already visible on chain, between about 6 and 7 million bitcoin depending on the count. All other coins are exposed only while a transaction spending them waits to be confirmed.

Sources

  1. 1.Magic state cultivation on a superconducting quantum processor — Science, October 15, 2026
  2. 2.Bringing the magic: Google breaks open a bottleneck in quantum computing — Google, October 15, 2026
  3. 3.Magic state cultivation on a superconducting quantum processor, preprint — arXiv:2512.13908, December 2025
  4. 4.Magic state cultivation: growing T states as cheap as CNOT gates — arXiv:2409.17595, September 2024
  5. 5.Securing Elliptic Curve Cryptocurrencies against Quantum Vulnerabilities: Resource Estimates and Mitigations — arXiv:2603.28846, March 2026
  6. 6.How to factor 2048 bit RSA integers with less than a million noisy qubits — arXiv:2505.15917, May 2025
  7. 7.Compact Shielded CSV: Post-Quantum, Private, Lightweight Client-Side Validation Blockchain — arXiv:2609.32015, September 2026
  8. 8.Shielded CSV: Private and Efficient Client-Side Validation — IACR ePrint 2025/068
  9. 9.Bitcoin Researcher Highlights 96-Byte Post-Quantum Transaction Proposal — The Crypto Times, October 14, 2026
  10. 10.Measuring Bitcoin's Quantum-Exposed Supply — Glassnode, May 20, 2026
  11. 11.Project Eleven Awards 1 BTC Q-Day Prize for Largest Quantum Attack on Elliptic Curve Cryptography to Date — Project Eleven, April 24, 2026
  12. 12.Rational Quantum Mechanics: Testing Quantum Theory with Quantum Computers — PNAS, March 16, 2026
  13. 13.Solving the Mysteries of Quantum Mechanics: Why Nature Abhors a Continuum — arXiv:2602.16382, February 2026

Not financial advice. CanoeBit publishes educational content only. Nothing here is a recommendation to buy, sell, or hold any asset.