Banks Are Running Out of Time on Post-Quantum Cryptography Migration

The cryptography protecting a wire transfer today is not the cryptography that will need to protect it in 2035 — and for financial institutions, the deadline to swap one for the other is closer than most risk committees have budgeted for. In August 2024, NIST finalized three post-quantum cryptography standards — ML-KEM (based on Kyber) for key exchange, and ML-DSA (Dilithium) and SLH-DSA (SPHINCS+) for digital signatures — closing an eight-year public evaluation process. The US National Security Agency's CNSA 2.0 timeline requires national security systems to support these algorithms by 2027 and reach full exclusive use by 2033. Banks aren't bound by CNSA 2.0 directly, but regulators in the US, EU, and UK are converging on similar windows, and the financial sector is, by its own admission, behind.
The urgency isn't that a working quantum computer capable of breaking RSA-2048 or elliptic-curve cryptography exists today — it doesn't. It's "harvest now, decrypt later": an adversary with resources can record encrypted traffic now and decrypt it once a sufficiently capable quantum computer exists, which most serious estimates place somewhere in the 2030s, with a wide error bar. For data with a short shelf life, that's not a threat. For a bank, it's the opposite. Loan agreements, KYC records, trade settlement data, and inter-bank communications carry confidentiality requirements that run 10, 20, even 40 years. Data encrypted today with RSA or ECC and intercepted today is a liability the day a quantum computer capable of Shor's algorithm at scale comes online — regardless of when that happens.
Why the migration is genuinely hard, not just slow
The hardest part isn't choosing an algorithm — NIST did that. It's that post-quantum algorithms have fundamentally different performance characteristics than the RSA and ECC primitives wired into thirty years of financial infrastructure. A Kyber-768 public key runs about 1,184 bytes versus 32 bytes for X25519 — roughly 37 times larger. ML-DSA signatures are similarly bulkier than ECDSA. For a retail banking app doing one TLS handshake, that's noise. For a high-frequency trading system doing millions of handshakes a day, or a hardware security module designed around fixed-size key slots from the 2010s, that size increase means re-architecting network buffers, re-certifying HSMs, and in some cases replacing hardware that was never designed to hold a kilobyte-plus key.
Then there's the certificate and PKI problem. Every TLS certificate, every code-signing certificate, every HSM-backed root of trust in a bank's infrastructure was issued under RSA or ECC assumptions. Migrating means running hybrid certificates — classical plus post-quantum algorithms combined, so a break in one doesn't compromise the connection — across every system that terminates TLS, which for a large bank is thousands of endpoints: core banking platforms, SWIFT gateways, ATM networks, mobile app backends, third-party vendor integrations that the bank doesn't fully control. Hybrid mode is the recommended transition path precisely because it hedges against both a flawed new algorithm and a surprise quantum breakthrough, but it also means running two cryptographic systems in parallel for years, with double the key management overhead.
What's actually happening versus what's announced
Large banks have run PQC pilots — HSBC and others have publicly discussed quantum-secured transactions and PQC-TLS trials on specific corridors. Pilots are not migrations. A pilot proves an algorithm works on one link between two consenting parties who both upgraded. A migration means every counterparty, every legacy system, every vendor's API has to support the same hybrid handshake, and in finance, counterparties number in the thousands and vendor upgrade cycles are measured in years, not sprints. SWIFT's own messaging network and the correspondent banking relationships layered on top of it are a coordination problem an order of magnitude larger than any single bank's internal infrastructure.
Regulators are starting to push rather than just recommend. The EU's cybersecurity agency ENISA and the UK's NCSC have both published PQC migration timelines urging critical infrastructure — which explicitly includes financial services — to inventory cryptographic assets now and complete migration of the highest-risk systems well before 2030. "Inventory now" is the unglamorous first step most institutions haven't finished: knowing exactly which of your systems use which algorithm, in which library version, with which key sizes, across every internal and third-party system, is harder than it sounds for an organization that has been acquiring other banks and inheriting their tech debt for two decades.
What this means for anyone running infrastructure with long-lived confidentiality needs
You don't need to be a bank to face this problem — it applies to anyone storing data that needs to stay confidential for a decade or more: healthcare records, legal documents, long-term IP filings. The practical steps are the same regardless of sector: run a cryptographic asset inventory first, before picking vendors or algorithms — you can't migrate what you haven't mapped. Prioritize hybrid classical-plus-PQC deployment over a hard cutover, since it's the only approach that doesn't bet everything on one algorithm family being flawless. Budget for hardware refresh, not just software patches — HSMs and smart cards with fixed key-size assumptions are often the actual bottleneck, not the TLS stack. And treat "harvest now, decrypt later" as a reason to start on data with the longest confidentiality horizon first, rather than waiting for a compliance deadline to force sequencing that should be driven by actual data sensitivity.