A ctlplne studio product
22 October 2026 · 5 min read

Harvest now, decrypt later: the quantum attack that starts before the computer exists

Someone can copy your encrypted traffic today and read it once a large enough quantum computer exists. You will not see it happen, and you cannot undo it. The fix is a setting most of your software already supports, and the usual reason it is off is a configuration written to be more secure.

Post-quantumMachine identityPKI fundamentals

What gets harvested

An encrypted session is only as private as the key exchange at its start. Record the whole conversation, and a future quantum computer could recover the session key from the recorded handshake, and with it every byte that followed. The US cyber agencies call this a catch now, break later or harvest now, decrypt later operation, aimed at data that still needs protecting years from now.1 The certificate is a different story:

Part of the connectionBreakable from a recording?What a future quantum computer gets
Key exchange: ECDHE or X25519, or RSA key transport in old TLS 1.2 setupsYesThe session key, and every byte of the recorded session
Certificate signature: ECDSA or RSANoNothing from the past. A forged signature only works on a live connection made after the computer exists

That is why hybrid post-quantum key exchange already protected more than half of the human traffic Cloudflare saw by the end of 2025, while public certificates are still classical.2,3

Does it matter to you?

Michele Mosca reduced it to three numbers:4 x, how many years your data must stay secret; y, how many years your migration takes; and z, how many years until a large enough quantum computer exists. If x + y > z, data you send before the migration finishes is exposed for the overlap.

Mosca's inequality on a timeline from today: the migration takes y years, data sent before it finishes must stay secret for x more years, and a large quantum computer arrives at z. The part of x that runs past z is exposed. z · a large quantum computer exists (date unknown) exposed y · your migrationx · data must stay secret todayyzy + x
Nobody controls z. You control y, how fast you migrate, and for credentials you also control x: it is how long they stay valid.

Nobody knows z. NIST's draft transition plan disallows quantum-vulnerable key exchange in federal systems after 2035.5 If your data must stay secret for ten years, health records, legal files, designs, and you treat 2035 as z, the inequality already fails.

The secrets nobody counts

Machine credentials cross the wire all day: API keys in headers, database passwords, bearer tokens, keys pushed out by deployment tools. For a credential, x is simply how long it stays valid. A key rotated every ninety days is worthless long before anyone decrypts the session that carried it. A service-account key from 2019 that still works in 2035 is the prize. Rotation is the one defense that also covers the sessions you cannot upgrade.

The fix, and the habit that turns it off

The fix is hybrid key exchange. X25519MLKEM768 runs classical X25519 and ML-KEM-768 from FIPS 203 side by side, so the session stays private if either one holds.6 Go 1.24, OpenSSL 3.5, Chrome 131 and iOS 26 all use it by default.7,8,9,3

The trap is hardening. Go enables the hybrid only when a program leaves its key-exchange group list unset,7 and nginx's ssl_ecdh_curve defaults to auto, OpenSSL's own list, which any explicit list replaces.10 A group list pinned for an old audit now pins you out of post-quantum. We found this in our own code: trstctl's listeners pin X25519, P-256 and P-384, so they negotiate classical key exchange. Adding the hybrid group is next.12

One caution before you switch it on everywhere: the hybrid key share adds more than a kilobyte to the first message, and some servers and middleboxes fail when it spans more than one packet.7,11 Test, then switch.

Check an endpoint in a minute

You need OpenSSL 3.5 or later; the LibreSSL built into macOS refuses the option. Offer only the hybrid group:

openssl s_client -connect example.com:443 -tls1_3 -groups X25519MLKEM768 -brief </dev/null

Support shows as Negotiated TLS1.3 group: X25519MLKEM768. Without it, the handshake fails with SSL alert number 40. In a successful run, the signature line a few rows up still reads ecdsa_secp256r1_sha256: post-quantum key exchange, classical certificate. Then check everything behind the edge: origins, internal service calls, appliances. Your CDN has probably upgraded the public side; nothing behind it upgrades itself.

Hybrid key exchange cannot protect what was already recorded, which is the argument for switching this quarter. It does not make certificates post-quantum either; that is a separate, slower migration.

trstctl's part is the inventory and the rotation. Its discovery finds the TLS endpoints and certificates you actually have, its cryptographic inventory flags each quantum-vulnerable key it scans with its FIPS 203, 204 or 205 replacement, and its secret store prefers short-lived, dynamic credentials that cannot be hoarded, issuing expiring API keys and database credentials on demand.12

Sources

  1. CISA, NSA and NIST, Quantum-Readiness: Migration to Post-Quantum Cryptography, 21 August 2023: "catch now, break later or harvest now, decrypt later" operations against data with a long secrecy lifetime.
  2. Cloudflare, State of the post-quantum Internet in 2025, 28 October 2025: why key agreement is urgent and signatures less so; more than half of human traffic protected.
  3. Cloudflare, The 2025 Cloudflare Radar Year in Review: post-quantum encrypted traffic from 29% to 52% during 2025, driven by iOS 26.
  4. Michele Mosca, Cybersecurity in an era with quantum computers: will we be ready?, IACR ePrint 2015/1075; IEEE Security & Privacy, 2018.
  5. NIST, IR 8547, Transition to Post-Quantum Cryptography Standards, initial public draft, 12 November 2024: quantum-vulnerable key establishment disallowed after 2035.
  6. NIST, Announcing Approval of Three Federal Information Processing Standards for Post-Quantum Cryptography, 13 August 2024: FIPS 203, ML-KEM.
  7. Go, Go 1.24 Release Notes, crypto/tls: X25519MLKEM768 on by default when Config.CurvePreferences is nil; servers that mishandle large records.
  8. OpenSSL, OpenSSL 3.5 release notes, 8 April 2025: default groups prefer hybrid post-quantum groups; default key shares X25519MLKEM768 and X25519.
  9. Google Security Blog, A new path for Kyber on the web, 13 September 2024: ML-KEM in Chrome 131.
  10. nginx, ssl_ecdh_curve: the default auto uses OpenSSL's built-in list; an explicit list sets the server's curves.
  11. tldr.fail: post-quantum ClientHello messages exceed a single packet, and servers that read them in one call fail.
  12. trstctl, Crypto agility and post-quantum: hybrid key exchange on the listeners is next. Documentation: Secrets, Discovery and inventory.