Zenpanik
Two clocks: quantum breaks the math, AI breaks everything around itBy Pavel Nikolov, November 11, 2026

Part 2: The quantum clock

Part 2 of 5 in a deep dive on how quantum computing and AI are breaking cryptography from two directions. Start with Part 1 if you are new to the series.

What a quantum computer breaks

A cryptographically relevant quantum computer threatens one family of cryptography and leaves another almost untouched. Factoring and the discrete logarithm are both problems of finding the period of a function such as f(x) = ax mod N. No fast classical method for that is known, while Shor's algorithm finds the period in polynomial time on a quantum computer. Every public-key scheme in wide use rests on one of those two problems: RSA, Diffie-Hellman, and the elliptic-curve schemes behind TLS key exchange, code signing, passkeys and every major blockchain. The other family, symmetric ciphers and hash functions, faces only Grover's algorithm, which speeds up brute-force search quadratically: searching N possibilities takes about √N steps. A 256-bit key therefore keeps the equivalent of 128 bits of security, which is why AES-256 and SHA-256 remain sound.

Symmetric encryption stays. Public-key algorithms do two jobs: key exchange, in which two parties agree on a secret over an open channel, and signatures, which prove who sent a message. Every public-key algorithm gets replaced, using the standards NIST published on August 13, 2024: ML-KEM (FIPS 203) for key exchange, ML-DSA (FIPS 204) and SLH-DSA (FIPS 205) for signatures, and the stateful hash-based schemes LMS and XMSS from SP 800-208 for firmware. ML-KEM and ML-DSA rest on a different hard problem, finding short vectors in lattices with hundreds of dimensions, for which no efficient quantum algorithm is known. SLH-DSA, LMS and XMSS rely only on hash functions. In March 2025 NIST also selected HQC, a key-exchange scheme built on error-correcting codes, as a backup to ML-KEM in case the lattice assumptions behind ML-KEM fail.

PrimitiveWhere fintechs use itQuantum impactReplacement
RSATLS certificates, code signing, banking PKI, HSM key wrappingBroken by ShorML-KEM for key transport, ML-DSA or SLH-DSA for signatures
ECDH (X25519, P-256)TLS and messaging key exchangeBroken by ShorML-KEM, deployed today as a hybrid with X25519
ECDSA, EdDSA, SchnorrBlockchains, passkeys, API signing, firmwareBroken by ShorML-DSA, SLH-DSA; LMS or XMSS for firmware
AES-256Data at rest, card data, backupsWeakened modestly by GroverKeep; prefer 256-bit keys
SHA-256, SHA-3Hashing, mining, address commitmentsWeakened modestly by GroverKeep

RSA has been public since 1977 and elliptic-curve cryptography since 1985, and neither has been broken at deployed key sizes. That record is real, and it is also the record of Taleb's turkey. In The Black Swan, a turkey fed every day for a thousand days grows more confident in the farmer with each meal, right up to the Wednesday before Thanksgiving. Taleb's line for what happens next is that the turkey “will incur a revision of belief.” Fifty years of classical attacks carry no information about Shor's algorithm, because the threat comes from a kind of machine that has never run against these keys.

The estimate falls faster than the hardware rises

A qubit is the quantum counterpart of a bit, and today's physical qubits are noisy. The estimates below assume qubits that fail about once in a thousand operations, so hundreds of them have to work together, through error correction, as one reliable logical qubit. That overhead is why the counts run into the millions.

In 2019 Craig Gidney and Martin Ekerå estimated that factoring a 2048-bit RSA key would take about 20 million noisy physical qubits running for eight hours. In May 2025 Gidney brought that below one million, with under a week of runtime and the same assumptions about hardware quality, so the whole twentyfold gain came from better algorithms and error-correction layouts. In 2026 the estimates kept falling. Iceberg Quantum's Pinnacle design put RSA-2048 below 100,000 physical qubits running for about a month, and a team from Oratomic and Caltech described a design with roughly 11,000 to 13,000 atoms that would run for decades.

On October 28, Alice & Bob posted a preprint estimating that 22,593 physical cat qubits could factor RSA-2048 in 3.14 days. Cat qubits are a design that suppresses one of the two kinds of quantum error in the hardware itself, which cuts the error-correction overhead. The paper has not been peer reviewed, and the largest chip on which the company has reported error-correction results has 12 cat qubits. Elliptic curves need less, because their keys are much shorter. In March, Google Quantum AI estimated that 256-bit elliptic-curve keys, the kind that secure Bitcoin, Ethereum and most TLS key exchange, could fall to fewer than 500,000 physical qubits in a few minutes, about 20 times fewer than the previous estimate.

101001,00010,000100,0001 million10 million100 million201920202021202220232024202520262027Quantinuum Helios-1: 98 physical qubitsGidney and Ekerå, 20 million, 8 hoursGidney, under 1 million, under a weekIceberg Pinnacle, under 100,000, about a monthAlice & Bob, 22,593 cat qubits, 3.14 daysOratomic and Caltech, about 11,000 to 13,000 atoms, decades of runtimeGoogle, ECC-256: under 500,000, minutesFault-tolerant runs on Helios-1 used up to 8 logical qubits. RSA-2048 needs about 1,400.Physical qubits, log scale. Each estimate assumes different hardware, error rates and runtimes. None has been built.
RSA-2048 estimateRSA-2048, very long runtimeElliptic curves (ECC-256)Largest high-accuracy machine
Figure 3. Physical qubits needed to break RSA-2048 by published estimate, against the qubits in Quantinuum's Helios-1. Sources: Gidney, PostQuantum.com, Alice & Bob, Google Quantum AI, PostQuantumSecurity tracker, Quantum Computing Report.

The hardware gap is still wide. Quantinuum's Helios-1 has 98 physical qubits. In a September 2026 benchmark by Sandia, Quantinuum and NVIDIA, its fully fault-tolerant runs used up to 8 logical qubits, and RSA-2048 needs about 1,400. That distance, from 8 reliable qubits to about 1,400, is why nobody can break RSA today. The estimates, meanwhile, have dropped about three orders of magnitude in seven years. Hardware quality is improving too. On October 19, Quantinuum reported a logical fidelity of 99.9995% on Helios, about five errors per million operations, with logical errors falling exponentially as the code grew. The company targets fault-tolerant computing by 2029.

Expert views on timing differ widely. Justin Drake, an Ethereum Foundation researcher and co-author of Google's paper, put at least a 10% chance on a quantum computer recovering a bitcoin key from an exposed public key by 2032. According to Cloudflare, IBM Quantum Safe's CTO cannot rule out targeted attacks on high-value keys as early as 2029. Dan Boneh of Stanford, another co-author, considers a capable machine before 2035 possible but unlikely at current funding levels. I use 2029 to 2035 as the range in the charts below.

Harvest now, decrypt later

Signatures and encryption fail differently. A forged signature needs the quantum computer to exist at the moment of forgery, so signatures made today stay trustworthy until then. Encryption fails backwards in time. An adversary who records encrypted traffic today can store it and decrypt it once a capable machine exists. For data that must stay secret for years, the exposure has already started.

Michele Mosca turned this into a planning rule in 2015. Let x be the number of years your data must stay confidential, y the years your migration will take, and z the years until a cryptographically relevant quantum computer exists. If x plus y is greater than z, information protected with today's algorithms will become readable while it is still meant to be secret. With x = 10 years and y = 6 years, the data is safe only if z is more than 16 years.

Range of expert views, 2029 to 203520262030203520402045205020552060Payment session dataexample: x is 1 yearsecret until 2033KYC and AML recordsexample: x is 10 yearssecret until 2042Loans, pensions, health dataexample: x is 25 yearssecret until 2057y: six-year migration (assumed)Schematic, not data. Grey = migration time (y). Amber = how long the last classically protected data must stay secret (x).
Figure 4. Mosca's inequality for three kinds of fintech data, assuming a six-year migration from 2026. Session data is exposed only if quantum arrives early in the range (light amber). Customer records and long-dated contracts stay secret past the whole range (amber). The secrecy lifetimes are examples, not regulatory periods.

Google separates the two jobs in its March post. Encryption faces store-now-decrypt-later attacks today. Signatures are a future threat, but they have to be migrated before a quantum computer exists, because forged signatures would let attackers impersonate trusted parties and distribute malicious software. Google moved authentication up its own priority list for that reason.

Why crypto assets are first in line

Blockchains concentrate the risk. A coin belongs to whoever can produce a signature that matches the public key the coin is locked to, and recovering the private key from the public key is the elliptic-curve discrete logarithm problem, which Shor's algorithm also solves. Elliptic-curve keys are much smaller than RSA keys of comparable strength, so a smaller quantum computer can break them, and on a public chain the inputs to the attack stay readable forever. Google's paper estimates that a capable machine could derive a bitcoin private key in about nine minutes once the public key is exposed, close to the ten-minute block interval. A transaction reveals its public key when it is broadcast and is confirmed about ten minutes later, so an attacker that fast could redirect a payment while it waits.

Counts of exposed coins depend on method. Google's paper puts about 6.9 million BTC, a third of supply, in wallets whose public keys are already visible. The Quantum Alliance counts about 6.04 million. The BIP-361 draft puts the share of bitcoin with a revealed public key above 34% as of March 1, 2026.

Here the quantum clock meets the hardware wallets from my last deep dive. A hardware wallet keeps the private key inside a secure element and never exports it. Once the public key is on-chain, a quantum attacker derives the private key from public data and never touches the device. Many addresses are a hash of the public key, and no quantum algorithm reverses a hash, so the key stays hidden until the owner first spends from the address. What protects a coin is whether its public key has been revealed, and that depends on the output type and on how the owner has used the address.

Output typeWhere the public key appearsExposed at restExposed while spending
P2PK (early coins, including Satoshi-era)In the output itself
P2PKH or P2WPKH, reused after a spendRevealed by the first spend
P2TR (Taproot) key pathIn the output itself
P2PKH or P2WPKH, never spent fromHidden behind a hash until spent
P2MR (BIP-360 draft)Never in the output
ExposedExposed while the spend waits in the mempool, or depending on the script usedNot exposed
Figure 5. Where a bitcoin public key becomes visible, by output type. A hardware wallet protects the private key in every row; only the output type and the owner's habits protect the public key. Sources: BIP-360, Google Quantum AI.

A BIP, or Bitcoin Improvement Proposal, is the formal route for changes to the protocol. BIP-360 was merged into the Bitcoin BIPs repository as a draft on February 11, 2026. Its Pay-to-Merkle-Root outputs remove Taproot's key-path spend, so the public key never sits in the output. It does nothing for coins already in vulnerable outputs; their owners have to move them. Nobody has settled whether a network should ever freeze dormant coins that cannot be migrated, and no technical proposal can settle it alone. Google's paper recommends that holders stop exposing and reusing vulnerable addresses, and sets out policy options for abandoned coins.

BIP-361, a draft proposal that builds on BIP-360, sets out a phased sunset: about three years after activation, sending to vulnerable addresses would stop, and a later phase would freeze coins left in them. On October 7, 2026, Europol published two reports that name wallets with exposed public keys as the main quantum exposure. Europol found no cryptographic remedy for keys already exposed and recommends pre-emptive migration and crypto-agility. Its second report covers harvest-now-decrypt-later risk to data that stays sensitive for years.

The regulatory clock

Regulators have turned the uncertain date into fixed deadlines. Under the NSA's CNSA 2.0, new acquisitions for US national security systems must be compliant from January 1, 2027. NIST finalized IR 8547 on October 27, 2026. Under it, quantum-vulnerable public-key algorithms are deprecated after 2030 and disallowed after 2035. Deprecated means still permitted but discouraged; disallowed means federal systems may no longer use them. Executive Order 14412 of June 22, 2026 requires federal high-value systems to use post-quantum key establishment by December 31, 2030 and post-quantum signatures by December 31, 2031, and OMB memorandum M-26-15 sets five phases from discovery in 2026 to full migration in 2035. The EU's coordinated roadmap asks member states to start by the end of 2026, migrate high-risk systems by 2030 and finish by 2035.

Financial services have their own reference point. In January 2026 the G7 Cyber Expert Group, which advises G7 finance ministers and central bank governors, published a roadmap for the financial sector. It sets no regulatory expectations, but it lays out phases from awareness and inventory to migration and validation, cites 2035 as the widely accepted completion target, and recommends building crypto-agility into transition plans.

Range of expert views, 2029 to 20352024202620282030203220342036NISTFIPS 203 to 205DeprecateDisallowNSA CNSA 2.0New purchasesPhase out legacyMost completeUS federal civilianDiscoveryKey exchangeSignaturesFullEU roadmapStartHigh-risk systemsCompleteG7 financial sectorRoadmapComplete (target)Google, CloudflareOwn systems done
Figure 6. Post-quantum deadlines against the range of expert views. Most mandates land inside the range. Sources: NIST, NSA, EO 14412, OMB, EU NIS Cooperation Group, G7 Cyber Expert Group, Google, Cloudflare.

Against Mosca's rule, the mandates work if quantum arrives late in the range. If it arrives early, data recorded today and keys exposed today are lost, whatever their compliance status. The quantum clock comes with a rough date and a known fix. The migration it forces is a decade of change, and Part 3 shows how AI attacks systems while they change.

Continue to Part 3: The AI clock