Join our Newsletter — 33% off our NHI Course
Home› Guides› Post Quantum KEMs vs Signatures: why harvest now…
Guide Data Security

Post Quantum KEMs vs Signatures: why harvest now decrypt later sets the order

PDF for newsletter subscribers · free to read online

← All guides
By NHI Mgmt Group, edited by Lalit Choda Updated 14 November 2026 17 min read
On this page

Post-quantum key encapsulation and post-quantum signatures defend against different attacks, so treating them as one migration wastes effort on the wrong systems first. A key encapsulation mechanism (KEM) such as ML-KEM protects the secrecy of a session, and that secrecy can be broken years after the traffic was captured. A signature such as ML-DSA protects authenticity, and a forged signature only matters once an attacker can actually forge it. This guide compares the two side by side: threat model, architecture, failure modes, controls and timelines, then gives a decision checklist and a worked scenario. It is written for security architects, PKI owners and migration leads who already know the algorithms and need to decide what moves first. For algorithm detail, see the ML-KEM Guide and the ML-DSA and SLH-DSA Post-Quantum Signatures Guide.

Key takeaways

  • Harvest now, decrypt later only hits confidentiality. Recorded traffic protected by classical key exchange can be decrypted once a cryptographically relevant quantum computer (CRQC) exists, which is why NIST IR 8547 says it expects to prioritise migration to quantum-resistant key establishment.
  • Signatures are safe until the day they are not. IR 8547 states that, unlike encryption, an authentication system remains secure against harvest now, decrypt later; the risk is forgery after a CRQC, so urgency follows how long a signature or trust anchor must be relied on.
  • KEMs change at the endpoints, signatures change across the trust chain. Hybrid key exchange needs two peers to agree a new group; a new signature algorithm needs every verifier, CA, HSM and bootloader in the chain to understand it.
  • The dates are close but not identical. IR 8547 (final, 27 October 2026) deprecates quantum-vulnerable signatures and key establishment after 2030 and disallows them after 2035, while asking high value assets to adopt post-quantum key establishment by 31 December 2030 and signatures by 31 December 2031.
  • Start signatures early where the verifier cannot be changed. Firmware roots, long-lived CAs and devices that outlive 2035 need post-quantum verification designed in now, even though the forgery threat is later.

The short answer

Migrate key establishment first for anything carrying data that must stay secret beyond the likely arrival of a CRQC, because every session recorded today is already exposed. Migrate signatures on a trust chain timeline: start with the anchors that are hardest to replace (hardware roots of trust, private root CAs, code signing keys whose signatures are verified for years), and leave short-lived authentication such as TLS handshake signatures and workload certificates until CA, library and browser support is settled. IR 8547 captures both halves: it expects to prioritise key establishment, and it also says that while signatures are not vulnerable to harvest now, decrypt later, transitioning to post-quantum signatures is also urgent. The Post-Quantum Cryptography Migration Guide covers the programme as a whole; this page is only about the ordering decision.

Side by side

DimensionPost-quantum KEM (ML-KEM, FIPS 203)Post-quantum signature (ML-DSA FIPS 204, SLH-DSA FIPS 205, LMS/XMSS)
Property protectedConfidentiality of a shared secret and everything encrypted under itAuthenticity and integrity: who signed, and that nothing changed
Quantum attackRecover the session key from recorded handshakes, retroactivelyDerive a private key from a public key, then forge new signatures
When harm occursAny time after capture, including for traffic captured years before a CRQCOnly after a CRQC exists and only while the forged signature is still trusted
Who must changeThe two endpoints of a protocol (client and server, initiator and responder)Signer, every verifier, the CA hierarchy, revocation, HSMs and any embedded root of trust
Typical first deploymentHybrid TLS 1.3 group X25519MLKEM768, SSH mlkem768x25519-sha256, IKEv2 additional key exchangeFirmware signing with LMS or XMSS, lab or private PKI with ML-DSA certificates (RFC 9881)
Hybrid formConcatenated shared secrets (SP 800-56C composite technique, as IR 8547 describes)Dual or composite signatures where all components must verify (RFC 10073)
Size impactX25519MLKEM768 client share 1,216 bytes, server share 1,120 bytes (RFC 10024)ML-DSA-65 public key 1,952 bytes, signature 3,309 bytes (RFC 9881), against 64 bytes for an ECDSA P-256 signature
RollbackEasy: drop the hybrid group, peers negotiate a classical groupHard: issued certificates, signed firmware and published packages persist
Main standard dates (IR 8547)Classical DH, ECDH and MQV deprecated after 2030, disallowed after 2035RSA, ECDSA and EdDSA deprecated after 2030, disallowed after 2035

KEMs in depth: threat model, architecture and failure modes

Threat model

An adversary records encrypted traffic now, stores it cheaply, and decrypts it once a CRQC can solve the discrete logarithm problem behind the ECDHE key share. Nothing in the session reveals the theft, and no later change to your systems can protect what was already captured. IR 8547 names TLS as an early priority because it protects data in transit and is a likely collection target, and it singles out interactive protocols such as TLS and IKE. The OpenSSH project makes the same argument for SSH, stating that traffic sent today is at risk of decryption unless post-quantum key agreement is used. The exposure window for any dataset is its secrecy lifetime plus your migration time; if that extends past your estimate for a CRQC, the data is already at risk. The Harvest Now Decrypt Later Risk Assessment Template gives a structured way to score this per data flow.

Architecture in words

In TLS 1.3 the client sends a key share for a hybrid named group; RFC 10024 (Proposed Standard, August 2026) defines X25519MLKEM768 (codepoint 0x11EC, marked recommended), SecP256r1MLKEM768 (0x11EB) and SecP384r1MLKEM1024 (0x11ED). The server encapsulates to the ML-KEM public key, returns a ciphertext alongside its classical share, and both sides feed the two shared secrets into the key schedule. Only the handshake changes: certificates, the cipher suite and application data are untouched. In SSH, OpenSSH has used mlkem768x25519-sha256 by default since release 10.0 (April 2025), after offering sntrup761x25519-sha512 by default from 9.0 (April 2022). In IPsec, an additional key exchange carries the ML-KEM component, which is the IETF approach that the NSA CNSA 2.0 FAQ points to for IKEv2. See the Hybrid Post-Quantum TLS Key Exchange Guide and the Post-Quantum Readiness for VPNs and IKEv2 Guide.

Failure modes

  • Middlebox breakage: a client hello that grows by more than a kilobyte can exceed a single packet; older inspection devices and load balancers that expect a one-packet hello may reset the connection.
  • Silent downgrade: if a server or proxy does not offer the hybrid group, peers negotiate classical ECDHE without error, so the protection simply is not there.
  • Terminated in the middle: TLS inspection or a CDN re-encrypts to the origin with classical key exchange, so only the first hop is protected.
  • Long-term keys outside the handshake: data encrypted at rest under RSA or ECDH key wrapping (backup encryption, envelope encryption, S/MIME) carries the same retroactive exposure and is easy to miss.

Signatures in depth: threat model, architecture and failure modes

Threat model

A signature cannot be recorded now and broken later in a way that hurts past sessions: a TLS handshake signature verified in 2026 authenticated a connection that is long over. The attack begins once a CRQC can recover a private key from a published public key. From then, the attacker can issue certificates from a classical CA, sign malware with a classical code signing key, or mint tokens with a classical identity provider key. OpenSSH states that the only urgency for signature algorithms is ensuring that all classical signature keys are retired. The NCSC adds the trust chain point: quantum-secure authentication is only achieved once PKI migration is complete and traditional certificates have expired or been revoked. The Storm-0558 incident, where a stolen consumer signing key that should have been retired was used to forge tokens accepted for enterprise mail, shows what one trusted signing key can do in the wrong hands; see the Microsoft Storm-0558 Key Breach.

Architecture in words

A post-quantum signature has to be supported at every link. The CA must hold an ML-DSA or SLH-DSA key, typically in an HSM with validated firmware; certificate profiles must carry the new algorithm identifiers (RFC 9881 assigns OIDs for ML-DSA-44, ML-DSA-65 and ML-DSA-87 and allows only the pure variant in certificates); revocation lists and OCSP responses must be signed with it; and every relying party must hold the new root and run a library that verifies it. The NCSC describes a new post-quantum root that can run in parallel with the traditional PKI or cross-sign the old root. Where a smooth transition matters, composite certificates bind ML-DSA with RSA, ECDSA, Ed25519 or Ed448 into one key and one signature; RFC 10073 (Proposed Standard, November 2026) defines 19 composite algorithm identifiers and validates a signature only when every component verifies. For firmware, the NSA CNSA 2.0 FAQ lists LMS and XMSS for software and firmware signing, and IR 8547 notes that devices expected to remain in service after a CRQC may need to require quantum-resistant signatures on executable code because some devices cannot be updated after manufacture.

Failure modes

  • Verifier lag: a new root is issued but embedded devices, Java trust stores or partner systems cannot validate it, so the old root stays trusted indefinitely.
  • Classical key left trusted: post-quantum certificates are added but the classical root is never removed, so a future forger simply uses the old chain.
  • Size surprises: a chain of three ML-DSA-65 certificates adds about 16 kilobytes of keys and signatures, which can break constrained protocols, token headers and certificate fields with fixed limits.
  • Stateful key misuse: LMS and XMSS fail catastrophically if a one-time signature index is reused, for example after restoring an HSM from backup.

Why the migration order differs

DriverEffect on KEM timingEffect on signature timing
Retroactive exposurePulls migration forward: every day without it adds captured trafficNone: past authentications cannot be undone
Lifetime of trustSet by data secrecy lifetimeSet by how long a root, firmware key or signed artefact is relied on
Coordination neededTwo endpoints; libraries and browsers already ship hybrid groupsCA hierarchy, HSM vendors, every relying party; NCSC notes there is no agreed WebPKI approach yet
ReversibilityNegotiated per connection, so rollback is a configuration changeArtefacts persist for their lifetime, so mistakes are expensive
Standards maturityRFC 10024 hybrid groups published; NIST recommends ML-KEM-768 as default in IR 8547RFC 9881 and RFC 10073 published; browser and public CA programmes still forming
Hardware dependencyMostly software; HSMs rarely involved in session key exchangeCA and code signing HSMs, TPMs and boot ROMs must support the algorithm

The practical result is two overlapping tracks rather than a strict sequence. The KEM track is a broad, fast rollout across protocols you control. The signature track is a long, narrow programme that starts with design work on the hardest anchors and finishes when the last classical root is removed. The NCSC guidance reflects the same pattern, noting that services protecting confidentiality will be available soonest while certificate-based PKI, IoT and ICS protocols follow more slowly.

Timelines and what they ask for

SourceKey establishmentSignatures
NIST IR 8547 (final, 27 October 2026)DH, ECDH and MQV at 112-bit strength or above deprecated after 2030, disallowed after 2035; high value assets and high impact systems adopt post-quantum key establishment by 31 December 2030RSA, ECDSA and EdDSA deprecated after 2030, disallowed after 2035; high value assets and high impact systems adopt post-quantum signatures by 31 December 2031
NSA CNSA 2.0 FAQ (national security systems)ML-KEM-1024; new acquisitions CNSA 2.0 compliant from 1 January 2027; CNSA 2.0 algorithms mandated by 31 December 2031ML-DSA-87, with LMS or XMSS for software and firmware signing; same acquisition and mandate dates
NCSC PQC migration timelines (UK)Discovery, assessment and an initial plan by 2028, including any requirement to migrate long-lived hardware roots of trust; highest-priority migrations by 2031; all systems by 2035

The one-year gap in IR 8547 between key establishment (2030) and signatures (2031) for high value assets is the clearest official statement of the ordering. Track changes to these dates in the Post-Quantum Cryptography Deadlines Tracker.

Where identity fits

Most of the signature track is a machine identity problem. Workload certificates, service mesh mTLS identities, SSH host and user certificates, JWT signing keys for service accounts and code signing identities in CI pipelines are all signature keys bound to non-human identities. They also tend to be short-lived, which helps: a workload certificate that rotates daily can switch algorithm on the next issuance once the issuer and verifiers support it. The long-lived ones are the problem: identity provider token signing keys, SAML signing certificates with multi-year validity and static SSH host keys. Inventory them by owner and lifetime, and shorten lifetimes now so the eventual algorithm change becomes routine rotation. See the Post-Quantum Migration for NHIs Guide and the SSH Certificates vs X.509 Certificates comparison.

Decision guidance

  • Choose KEM first when: the data flow carries information that must stay confidential for more than about a decade (health, legal, state, long-lived intellectual property, credentials that are not rotated), or crosses networks you do not control.
  • Start signature design now when: the verifier lives in hardware or firmware that cannot be updated, the root CA has a validity beyond 2035, or signed artefacts (packages, firmware, archived documents) must verify for years.
  • Defer signatures when: certificates are short-lived, issued automatically and verified only by software you update regularly; keep the issuance pipeline agile and wait for library and CA support.
  • Use hybrid KEMs by default: commercial TLS and SSH defaults are hybrid, and IR 8547 describes hybrids as generally temporary. The CNSA 2.0 FAQ states that NSA does not require hybrid solutions, so national security system suppliers should check which configuration their customer expects.
  • Use composite or dual signatures selectively: they help where relying parties are mixed, at the cost of larger artefacts and more complex validation.
  • Small organisations: turn on hybrid key exchange where your platforms already offer it, ask suppliers for signature roadmaps and keep certificate lifetimes short. Running your own post-quantum PKI is rarely needed.
  • Large organisations and product vendors: run both tracks in parallel, build a lab PKI with ML-DSA and composite profiles now, and budget HSM refreshes for the signature track.

Worked scenario: a payments processor

A payments processor runs public APIs behind a CDN, internal service-to-service mTLS, a private CA with a root valid until 2040, SSH to a large server estate, and payment terminals with signed firmware and a ten-year field life.

  1. Classify flows by secrecy lifetime: cardholder data in transit and the VPN between data centres carry long-lived value; internal health checks do not.
  2. KEM track, first six months: enable X25519MLKEM768 on the CDN edge and confirm the CDN-to-origin hop also negotiates it; upgrade SSH clients and servers so mlkem768x25519-sha256 is negotiated; plan IKEv2 additional key exchange with the VPN vendor. Measure the share of handshakes using a hybrid group.
  3. Signature track, design phase: because the terminals stay in the field beyond 2035 and their boot verifier is in silicon, require post-quantum (LMS or ML-DSA) firmware verification in the next hardware generation and put it in the procurement contract now.
  4. Signature track, PKI: stand up a new ML-DSA root in a lab, test composite issuing CAs against the mesh and partner clients, and set the classical root to retire rather than renew.
  5. Defer: leave public TLS server certificates classical until public CA and browser programmes settle, while keeping certificate automation in place so the switch is a profile change.

Detection and measurement

  • Hybrid handshake share: percentage of TLS and SSH sessions negotiating a hybrid group, from load balancer, CDN and sshd logs, broken down by hop.
  • Classical fallback rate: sessions where both peers are capable but a classical group was negotiated, which points to a proxy or configuration gap.
  • Long-secrecy flows protected: share of flows scored high in the harvest now, decrypt later assessment that now use a post-quantum KEM.
  • Trust anchor inventory coverage: share of roots, issuing CAs, firmware and code signing keys with a recorded algorithm, expiry and replacement plan.
  • Verifier readiness: share of relying parties (devices, libraries, partners) tested to validate ML-DSA or composite signatures.
  • Classical anchors past 2035: count of classical roots or signing keys whose validity extends beyond 2035, trending to zero.

Framework mapping

FrameworkReferenceKEM trackSignature track
NIST SP 800-53 Rev 5SC-8 Transmission Confidentiality and IntegrityHybrid key exchange on linksHandshake authentication
NIST SP 800-53 Rev 5SC-12 Cryptographic Key Establishment and Management; SC-13 Cryptographic ProtectionApproved KEM and key scheduleApproved signature algorithms and key lifecycle
NIST SP 800-53 Rev 5SC-17 Public Key Infrastructure Certificates; SI-7 Software, Firmware, and Information IntegrityNot centralNew roots, certificate profiles, signed firmware and code
NIST SP 800-53 Rev 5CM-8 System Component Inventory; SA-22 Unsupported System ComponentsFind classical key exchangeFind devices that cannot verify new signatures
NIST CSF 2.0PR.DS-02 data in transit; PR.DS-01 data at rest; ID.AM-08 lifecycle managementPR.DS-02, PR.DS-01ID.AM-08 for roots and devices
ISO/IEC 27001:2022Annex A 8.24 Use of cryptographyKey exchange policySignature and key management policy

Common failure patterns

  • One programme, one queue: KEM and signature work sit in a single backlog ordered by system, so hybrid TLS waits behind a PKI redesign that will take years.
  • Edge only: the CDN or load balancer offers a hybrid group, but the hop to the origin, the service mesh or the backup replication link stays classical, leaving the most valuable traffic exposed.
  • Signatures left until 2034: teams read "no harvest now, decrypt later" as "no urgency" and discover that a boot ROM verifier or a 2040 root cannot be changed in time.
  • Adding without removing: new post-quantum or composite chains are issued but classical roots and signing keys stay trusted, so a future forger still has a path. Storm-0558 shows the cost of a signing key that outlived its retirement date.
  • Ignoring size: ML-DSA chains are deployed into protocols with fixed limits or into tokens carried in headers, and failures appear as intermittent errors rather than clear rejections.
  • Stateful keys treated as stateless: LMS or XMSS keys are backed up and restored like ordinary keys, risking index reuse.

Practitioner checklist

  • Split the migration into a KEM track and a signature track with separate owners, budgets and metrics.
  • Score every external and inter-site data flow by secrecy lifetime and move the high scores to hybrid key exchange first.
  • Enable X25519MLKEM768 in TLS 1.3 and mlkem768x25519-sha256 in SSH wherever your stack offers them, and verify every hop.
  • Monitor negotiated groups and alert on classical fallback between capable peers.
  • Inventory roots, issuing CAs, firmware, code and token signing keys with algorithm, expiry and verifier list.
  • Put post-quantum firmware verification (LMS, XMSS or ML-DSA) into procurement for any device expected to be in service after 2035.
  • Build a lab PKI with ML-DSA (RFC 9881) and composite (RFC 10073) profiles and test every relying party.
  • Shorten machine certificate and signing key lifetimes so the algorithm change becomes routine rotation.
  • Plan the retirement of every classical root and signing key, not only the issuance of new ones.

Standards and references

Related NHI Mgmt Group resources: Post-Quantum Cryptography Migration Guide · Crypto-Agility vs Post-Quantum Cryptography · Harvest Now Decrypt Later Risk Assessment Template · Post-Quantum Migration for NHIs Guide

PDF for subscribers

Take this guide with you

Every section, table and checklist in one PDF, with links to the sources and related guides. Subscribe to our weekly NHI & AI Identity Journal and we’ll email you a download link.

Free weekly newsletter

Subscribe to the NHI & AI Identity Journal

The latest on NHI and Agentic AI security – articles, research, breaches, news and events every week.

Bonus 33% off our NHI Course when you subscribe.

NHIMG Editorial Note
Researched and written by the NHI Mgmt Group editorial team using AI tools, under the editorial direction of Lalit Choda. Facts are drawn from the sources listed on this page and checked by our automated quality process. Last updated 14 November 2026. Spotted an error? Use our contact page.
NHI Mgmt Group, the independent authority on Non-Human Identity, IAM and Agentic AI security. nhimg.org

PDF Downloads for Newsletter Subscribers

Get the guide

Enter your details and we’ll email you a download link.

The full guide is free to read online. The PDF is a thank you for subscribing.

Fill in all fields marked * and tick the PDF box to continue.