The math on breaking Bitcoin just got worse
I work in the crypto self-custody space, and March 30, 2026 marks a clean before and after. That day Google Quantum AI published a paper (arXiv:2603.28846) that cuts the estimated hardware needed to break Bitcoin's keys by nearly 20 times and backs the result with a zero-knowledge proof. Nothing broke that day, but the time left to prepare got shorter.
The authors are Babbush, Zalcman, Gidney, Broughton, Khattar, Neven, Bergamaschi, Drake and Boneh. Their circuits solve the 256-bit elliptic curve discrete logarithm problem on Bitcoin's secp256k1 curve with about 1,200 logical qubits and 90 million Toffoli gates, or about 1,450 logical qubits and 70 million (arXiv:2603.28846). On superconducting hardware that comes to fewer than 500,000 physical qubits, which the paper calls "nearly a 20 fold reduction" over the previous best, Litinski's 2023 estimate (arXiv:2306.08585).
Google's own Willow chip, announced in December 2024, has 105 qubits (Google). Do not turn that into a countdown. The same paper argues that counting physical qubits fails to capture progress, which comes in jumps when a platform clears an error-correction threshold (Google Quantum AI whitepaper). In my view the paper is a data point, and it makes the eventual crisis easier to see.
TL;DR
Google's March 30, 2026 paper (arXiv:2603.28846) puts the cost of breaking a Bitcoin key at fewer than 500,000 physical qubits, nearly 20 times below the previous estimate. No quantum computer can break Bitcoin today. The paper counts a little over 1.7 million BTC in old P2PK outputs and about 6.9 million BTC in all outputs whose public key is already on-chain. A fresh bc1q address hides its public key until you spend. A Taproot bc1p address does not. Two draft proposals, BIP-360 and BIP-361, outline a migration path. NIST published its first post-quantum standards, one for key exchange (FIPS 203) and two for signatures (FIPS 204 and 205), on August 13, 2024 (NIST). SHA-256 and Proof of Work face no practical quantum threat. Prepare, but do not panic.
What the paper says
Read the author list once. Ryan Babbush, the first author, and most of his co-authors work at Google Quantum AI, the team that builds Google's quantum chips. Dan Boneh is a computer scientist at Stanford. Justin Drake is at the Ethereum Foundation in Zug (arXiv:2603.28846). Most of the authors build the hardware the paper is about.
Their contribution is a set of optimizations to Shor's algorithm for the elliptic curve discrete logarithm problem (ECDLP) under Bitcoin's signatures. The two circuits trade qubits for gates: 1,200 logical qubits and 90 million Toffoli gates, or 1,450 logical qubits and 70 million. With 10^-3 physical error rates and planar connectivity, the paper puts either one below 500,000 physical qubits and says they "can execute in minutes" (arXiv:2603.28846).
I like how the team handled disclosure. They published a zero-knowledge proof that the circuits exist, without publishing the circuits (Google Quantum AI whitepaper). Anyone can check the claim, and nobody gets the attack blueprint. In my view, that buys the community time.
A separate paper the same day, Cain et al. (arXiv:2603.28627), looks at neutral-atom qubits instead of superconducting ones. It finds Shor's algorithm possible with as few as 10,000 reconfigurable atomic qubits, and estimates a few days to break a key on the P-256 curve with 26,000 qubits. That is far slower than Google's minutes, because atom systems run slower error-correction cycles. Google's paper draws the same line between "fast-clock" and "slow-clock" machines. Two groups on two hardware paths both landed between 10,000 and 500,000 qubits. To me that matters more than either paper alone.
The same paper is also posted as IACR ePrint 2026/625.
Why the attack window matters
Bitcoin proves ownership with digital signatures: ECDSA, and since Taproot also Schnorr. Both rest on the elliptic curve problem, so both fall to Shor's algorithm (Google Quantum AI whitepaper). Your private key signs. Your public key verifies. A classical computer cannot work back from the public key to the private key. A large enough quantum computer can. For how keys and seed phrase security relate, the BIP39 guide covers the mechanics.
Your exposure depends on when, and whether, your public key is visible.
On-spend exposure is the time-sensitive case. When you broadcast a transaction, your public key sits in the mempool until a miner includes it. The Google paper estimates that a superconducting machine kept in a "primed" state could derive the private key about 9 minutes after it sees the public key, against Bitcoin's average block time of about 10 minutes (Google Quantum AI whitepaper). That overlap is why the authors say the first fast-clock machines would enable on-spend attacks.
At-rest exposure affects coins whose public key is already on-chain. The paper says P2PK outputs, the dominant type in the "Satoshi Era" of 2009 to 2010, hold a little over 1.7 million BTC, nearly 9% of all bitcoin (arXiv:2603.28846). It puts the total across all script types with an exposed key at about 6.9 million BTC. An attacker can hit these at any time, without waiting for a transaction.
For how Bitcoin addresses and key formats work, see Sending & Receiving Bitcoin and Technical Deep Dive.
The gap is real and the timeline is not infinite
Google's own machine is far from 500,000 qubits. Willow has 105 (Google). On the neutral-atom side, Cain et al. cite trapping arrays of more than 6,000 coherent qubits (arXiv:2603.28627). The Google authors warn that the final stages of the race may not be public, and that a public demonstration of Shor's algorithm on a 32-bit curve "should not be seen as a wake-up call" but as a sign that migration has already failed (Google Quantum AI whitepaper).
On March 25, 2026, Google set 2029 as its own deadline to migrate to post-quantum cryptography (Google). The post names progress in quantum hardware, error correction and factoring estimates as the reason, and moves authentication and signatures up the queue. When the company that builds the hardware sets its own deadline, I read that as a signal about the window.
NIST adds structure. Its first three post-quantum standards came out on August 13, 2024: FIPS 203 (ML-KEM, from CRYSTALS-Kyber) for key exchange, and FIPS 204 (ML-DSA, from CRYSTALS-Dilithium) and FIPS 205 (SLH-DSA, from SPHINCS+) for signatures (NIST). A NIST draft from November 2024, IR 8547, proposes deprecating ECDSA at 112-bit security after 2030 and disallowing ECDSA at every strength after 2035 (NIST IR 8547 ipd). That draft is still not final. Bitcoin does not answer to NIST. The wider ecosystem's schedule still gives you a useful floor for thinking about urgency.
What should register is velocity. The paper places its own result in a "general pattern of quantum algorithmic improvements", the same pattern that shrank the cost of breaking RSA-2048 (arXiv:2603.28846). In my view, attacks get better. They do not revert.
Which addresses show their public key
Bitcoin addresses do not carry equal exposure. The table below sorts the common types by what an attacker can see before you spend, using the Google paper's own classification (Google Quantum AI whitepaper).
| Script type | Address starts with | Public key on-chain before you spend? | At-rest risk |
|---|---|---|---|
| P2PK | (no address format) | Yes | Exposed |
| P2PKH | 1 | No, only a hash | None unless reused |
| P2WPKH | bc1q | No, only a hash | None unless reused |
| P2TR | bc1p | Yes, a tweaked key | Exposed |
Every type is exposed on-spend. Reuse breaks the "No" rows: once you spend from an address, its public key is on-chain, and the paper says coins still sitting there become as vulnerable as P2PK.
The paper says accounts believed to belong to Satoshi Nakamoto hold about 1 million BTC behind P2PK scripts, and that most of the 1.7 million P2PK bitcoin has not moved since 2009 (arXiv:2603.28846). The private keys may be lost. The owners may be unreachable.
That raises a policy question the paper takes on directly. It notes that many jurisdictions could classify taking abandoned coins as theft, argues the coins will be taken anyway if the protocol does not change, and discusses a legal framework for "digital salvage". The Bitcoin community has not settled the question. BIP-361 proposes first stopping sends to quantum-vulnerable addresses, then restricting old-signature spends behind a rescue protocol (BIP 361). I expect that debate to be loud.
The two Bitcoin proposals
Post-quantum work in Bitcoin predates this paper. BIP-360 was assigned its number on December 18, 2024 (BIP 360).
Both proposals are drafts. BIP-360, by Hunter Beast, Ethan Heilman and Isabel Foxen Duke, defines Pay-to-Merkle-Root (P2MR): Taproot with the key path spend removed, so no public key sits on-chain (BIP 360). It does not add post-quantum signatures yet; the Google paper calls it a security patch for the Taproot regression. BIP-361, with Jameson Lopp as first author, sets out two phases: about 3 years (160,000 blocks) after activation, no more sends to vulnerable scripts, then 2 years later, tighter rules on ECDSA and Schnorr spends (BIP 361). For the full analysis, read BIP-360 and BIP-361: Bitcoin's Quantum Upgrade Path.
The engineering constraints are real. Post-quantum signatures are larger. An ML-DSA-44 signature is 2,420 bytes (FIPS 204), against at most 73 bytes for a Bitcoin ECDSA signature including its sighash byte (BIP 66). That hits block space, fees and throughput. Any signature change needs a soft fork that the network has to adopt. Bitcoin upgrades move slowly on purpose. Developers can solve these problems, just not quickly.
For how Bitcoin's protocol architecture handles upgrades, see Technical Deep Dive.
What to do right now
Stop reusing addresses. Once you spend from an address, its public key is on-chain for good, and anything left on that address is exposed (Google Quantum AI whitepaper). The Bitcoin whitepaper already advised that "a new key pair should be used for each transaction" (whitepaper). Most modern wallets hand you a fresh address each time.
Pick the address type with the threat in mind. For coins you plan to hold for years, a fresh bc1q address hides the public key until you spend. A Taproot bc1p address shows a tweaked public key from the moment it receives coins, so the Google paper counts it as exposed at rest. The Bitcoin self-custody guide covers wallet choice in detail.
Check your old wallets. If you have been in Bitcoin since the early days and still hold coins in P2PK outputs or on an address you already spent from, moving them to a fresh bc1q address is the most concrete step available to you.
Follow BIP-360 and BIP-361 in the bitcoin/bips repository. When post-quantum proposals reach activation, the network will need to coordinate. Understand the timeline now and you are not scrambling when the upgrade ships.
For wallet security practices, see Wallets: Staying Secure. For privacy practices that also reduce quantum exposure, see Bitcoin Privacy.
References
- Babbush, R., Zalcman, A., Gidney, C., Broughton, M., Khattar, T., Neven, H., Bergamaschi, T., Drake, J., Boneh, D. "Securing Elliptic Curve Cryptocurrencies against Quantum Vulnerabilities: Resource Estimates and Mitigations." arXiv:2603.28846, March 30, 2026. Also IACR ePrint 2026/625. Full PDF: quantumai.google
- Cain, M. et al. "Shor's algorithm is possible with as few as 10,000 reconfigurable atomic qubits." arXiv:2603.28627, March 30, 2026.
- Litinski, D. "How to compute a 256-bit elliptic curve private key with only 50 million Toffoli gates." arXiv:2306.08585, June 2023.
- NIST. FIPS 203, 204 and 205, released August 13, 2024. NIST announcement. NIST IR 8547 (Initial Public Draft), November 2024. csrc.nist.gov
- Google. "Quantum frontiers may be closer than they appear," March 25, 2026. blog.google
What's next
New to Bitcoin? Start with Chapter 1. It covers the foundation you need to understand the rest of this article.
Already holding? Review your wallet setup in Wallets: Staying Secure and confirm you use a fresh bc1q address for each receive. That combination is your best protection today.
