Back to DashboardStrategyPhase 5 · Pilots & Migrationintermediate40 min

Hybrid & Composite Cryptography

Combine classical and PQC algorithms for defense in depth during the quantum transition.

Why this matters: Hybrid (classical + PQC) lets you deploy now without betting everything on algorithms still maturing in the field.

Start here: Choose KEM or Signatures and press Generate All Keys: classical, post-quantum and hybrid key pairs are generated side by side, with a table of key, ciphertext or signature sizes and gen time.

For your role

Developer / Engineer
Compare classical, PQC and hybrid key generation, run KEM encapsulation and hybrid signatures, generate the two root CAs and the hybrid X.509 formats, then the HPKE step, composed from PKCS#11 v3.2 mechanisms and shown beside the v3.3 draft composite KEM.
Security Architect
The certificate steps generate and compare the hybrid X.509 approaches with the IETF reference artifacts, and the signature-spectrum step compares concatenation, nesting and a local shared-binding model. Its checks do not implement published Silithium or prove non-separability; use versioned profiles, relying-party compatibility checks and implementation evidence for the production decision.
Researcher / Academic
The IETF reference artifacts in the certificate deep-dive and the three signature compositions with their non-separability properties are the material to check against the drafts.
Certification & Validation Engineer
Start with the “Hybrid in a validated module (IG D.S)” callout in the Learn tab, then Step 2 (hybrid KEM encapsulation and signing) and Step 5 (inspect the hybrid certificates against their IETF reference artifacts): they show which part is a tested primitive and which is composition built on top of it. Steps 6 and 7 (hybrid signature constructions, including the Silithium research proposal, and HPKE) are background for this role.
IT Ops / DevOps
Generate the classical and PQC root CAs and the hybrid certificate formats: which one a relying party that only knows ECDSA still validates is the property your fleet needs during migration.
Practice in the Simulation

Why Hybrid Cryptography?

The transition to post-quantum cryptography faces a fundamental dilemma: PQC algorithms are newer and less battle-tested than classical algorithms, yet the “” (HNDL) threat means waiting is dangerous. solves this by combining classical and PQC algorithms together.

Hybrid key establishment

A classical key exchange (for example X25519) and a post-quantum KEM (ML-KEM) run side by side, and a key combiner merges the two shared secrets. The session key stays secret as long as either component is unbroken, provided the combiner is a sound one (for example the TLS hybrid design or X-Wing).

Hybrid signatures

A message carries a classical and a post-quantum signature. A forgery requires breaking both algorithms, provided the verifier requires both signatures to be valid. A verifier that accepts either one lets an attacker strip the other.

Hybrid is a trade-off rather than a free upgrade. The next two sections set out what it risks compared with pure PQC, for key establishment and for signatures, and how national agencies, standards bodies and industry have decided. The rest of the lesson then builds from the primitives up: hybrid KEMs, hybrid signatures, and finally the certificates that carry them.

RFC 9794 standardizes terminology for hybrid schemes — distinguishing “composite” (single OID, both-must-verify) from “non-composite” (parallel independent algorithms) and establishing naming conventions such as “PQ/T” (Post-Quantum / Traditional).

Hybrid or Pure PQC: The Risks

Both choices carry risk. Pure PQC relies on algorithms with a shorter record of cryptanalysis and on new implementations. Hybrid adds a second algorithm and the machinery to combine the two. Each item below names the document that states it.

What pure PQC alone is exposed to

  • Cryptanalysis of young schemes

    SIKE and Rainbow, both NIST competition candidates, were broken with classical computers. In 2026 the BSI advised against Classic McEliece in new developments after new structural cryptanalysis, and noted that because it had only recommended hybrid use, a hybrid with Classic McEliece still keeps at least the security of the classical scheme.

    NÚKIB annex · BSI Classic McEliece notice
  • Bugs in new implementations

    Bernstein (2026) shows that a small change in ML-DSA software creates a flaw that interoperates with authentic ML-DSA, passes typical tests and is exploitable in 1 second on 1 laptop core. He estimates that far fewer Ed25519+ML-DSA keys than ML-DSA-only keys will be breakable, for years after the first quantum attack. The NCSC names implementation security as one reason to consider hybrid.

    Bernstein 2026 · NCSC white paper

Risks of hybrid KEMs

  • An unsound combiner

    Naïve constructions can be less secure than their components: a "concatenation only" hybrid session key is not secure against active adversaries, whatever the strength of each part. SP 800-227 approves specific key combiners (see Hybrid KEMs below).

    ETSI TR 103 966 · NIST SP 800-227
  • Fallback to classical-only

    Hybrids deployed for backwards compatibility can be open to downgrade attacks. If a client retries with a classical-only group after a hybrid attempt fails, an attacker who can make the post-quantum attempts fail obtains a classical-only session, even between two endpoints that both support hybrid (see TLS Basics).

    ETSI TR 103 966
  • Larger messages and denial of service

    The X25519MLKEM768 key share is 1,216 bytes instead of 32. In IKEv2, additional key exchanges before authentication increase the risk of denial-of-service attacks, because post-quantum KEMs can cost more than traditional key exchanges.

    ETSI TR 103 966
  • Protocol fit and validation

    HPKE's authenticated modes cannot be built on ML-KEM or on a hybrid that includes it. A FIPS 140-3 module may offer a hybrid KEM as an approved service only when it meets the five IG D.S conditions; the IG says X-Wing cannot yet meet them.

    ETSI TR 103 966 · FIPS 140-3 IG

Risks of hybrid signatures

  • Stripping one signature

    If a verifier accepts the hybrid when either component is valid, an attacker can remove the post-quantum signature and present the classical one. ETSI notes that such a backwards-compatible hybrid can be less secure than the post-quantum algorithm used alone. Requiring both signatures, or a non-separable construction (RFC 9955), closes this. The same applies one level up: a TLS server that keeps classical certificates for legacy clients can face a rollback in which an attacker strips the PQC or composite certificate; an individual draft proposes that clients remember the server’s PQC support.

    ETSI TR 103 966 · RFC 9955 · draft-sheffer-tls-pqc-continuity
  • More complex and less studied

    The NCSC notes that hybrid authentication can be significantly more complex than hybrid confidentiality, because both signatures must verify robustly, and that it has received significantly less research.

    NCSC white paper
  • The PKI has to carry both

    Hybrid authentication needs a PKI that issues both signature types, or two parallel PKIs. The NCSC prefers a single migration to a fully post-quantum PKI over an intermediate hybrid PKI. The NSA tells national security systems to stand up a parallel PKI and not to use novel mechanisms that carry several algorithms in one certificate.

    NCSC white paper · NSA CNSA 2.0 FAQ
  • Formats, size and approval

    No hybrid certificate format is universally accepted yet: composite certificates are not backward compatible, and GSMA PQ.03 considers it premature to recommend hybrid certificates. A composite ML-DSA-65 + ECDSA P-256 signature is about 3.4 KB, of which 64 bytes are the ECDSA part. The FIPS 140-3 IG has no section on composite signatures.

    GSMA PQ.03 · FIPS 140-3 IG

Risks common to both

  • Complexity and attack surface

    Hybrids add negotiation, error handling, API changes and testing. The NSA notes that more security products fail through implementation or configuration errors than through their algorithms, so added complexity can weaken security; the Dutch handbook notes that hybridisation may increase the attack surface.

    NSA CNSA 2.0 FAQ · Dutch PQC Migration Handbook
  • Interoperability and a second transition

    Every party must support all component algorithms and the same combination method. Removing the classical part later is a second transition; the NCSC asks that hybrid protocols allow a migration to a PQC-only future.

    NSA CNSA 2.0 FAQ · NCSC white paper
  • Proofs and labels

    For some formal security properties, hybrids may be no stronger, or even weaker, than their components; the NSA calls this a matter of significant debate. A product marketed as hybrid or composite is not necessarily quantum resistant.

    NSA CNSA 2.0 FAQ
  • Overhead

    Two algorithms run for each operation. When the hybrid is used once to set up a symmetric key for a connection, the Dutch handbook judges the added cost low compared with the whole connection.

    Dutch PQC Migration Handbook

Who Recommends Hybrid, and Who Does Not

Authorities weigh the same risks differently. France, Germany, the Netherlands and the EU certification group ask for hybrid during the transition; the NSA, the UK NCSC and the Czech NÚKIB accept post-quantum algorithms on their own. A recommendation applies only within its own scope, such as a national regulated perimeter or national security systems.

OrganizationPosition on hybridPure PQC on its ownSource
National agencies
France — ANSSIInsists strongly on hybrid wherever PQC is deployed, in the short and medium term. Mandatory inside the French regulated perimeter.Only for hash-based signatures (SLH-DSA, XMSS, LMS). May be revisited once the community has more hindsight.ANSSI PQC FAQ
Germany — BSITR-02102-1 currently recommends quantum-safe methods only in hybrid use with classical methods. Classical-only key agreement is recommended until the end of 2031.Not in the current version of the guideline.BSI TR-02102-1
Netherlands — AIVD, CWI, TNORecommends hybrid particularly for organisations that must deploy before mature implementations exist, for example when data is exposed to store-now-decrypt-later. Notes the overhead and the added attack surface.Not stated as a recommendation.PQC Migration Handbook
Czech Republic — NÚKIBAccepts both approaches for sensitive information in the near term: hybrid combinations, or standalone ML-KEM and ML-DSA at security level 5.Yes, at security level 5. Recommends standalone LMS and XMSS for firmware and software updates.NÚKIB annex
United Kingdom — NCSCHybrid where it is needed: interoperability, implementation security, or constraints of a protocol or system. Notes higher cost and complexity.Yes. Prefers a single migration to a post-quantum PKI over an intermediate hybrid PKI.NCSC white paper
United States — NSA (national security systems)Will not require hybrid for security purposes. Only hybrid solutions the NSA identifies count as CNSA 2.0 compliant, for example its IKEv2 profile, where message-size limits leave no straightforward standalone ML-KEM-1024 design.Yes: CNSA 2.0 algorithms on their own. A parallel PKI rather than several algorithms in one certificate.NSA CNSA 2.0 FAQ
United States — NISTSP 800-227: combining key-establishment schemes can be advantageous, for example during the migration, and it approves key combiners. It does not say how or when to migrate. FIPS 140-3 IG D.S lets a validated module offer a hybrid KEM under set conditions.Yes: ML-KEM and ML-DSA are approved on their own.NIST SP 800-227 · FIPS 140-3 IG
Canada — Cyber CentreITSP.40.111 recommends ML-KEM and ML-DSA for general use, and says classical key establishment and signatures used without a post-quantum scheme should be phased out by the end of 2035. It does not discuss hybrid constructions.Not stated either way.ITSP.40.111
Certification and standards bodies
EU — ECCG (EUCC certification)Agreed Cryptographic Mechanisms v2.0 recommends rolling out PQC in hybrid mode where confidentiality must hold for the long term, and agrees the CatKDF and CasKDF combiners. ENISA’s 2026 report on hybridization surveys standards and is explicitly not a recommendation.The v3 draft (not yet applicable) says lattice-based mechanisms should not be used standalone.ACM v2.0 · ACM v3 draft · ENISA report
ETSITS 103 744 specifies hybrid key exchanges; TR 103 966 describes the trade-offs of deploying them. Neither mandates hybrid.Not addressed.ETSI TS 103 744 · ETSI TR 103 966
IETFStandardizes both options: hybrid groups for TLS 1.3 (RFC 10024) and composite signatures in X.509 (RFC 10073), and pure ML-KEM groups for TLS 1.3 (still a draft). It does not choose between them for deployers.Specified alongside hybrid.RFC 10024 · RFC 10073 · draft-ietf-tls-mlkem
GSMA (telecom)PQ.03 finds it premature to recommend hybrid certificates while the community evaluates their feasibility and necessity.Not addressed.GSMA PQ.03
Industry practice and research
Google Chrome, Cloudflare, AWSDeploy hybrid key exchange (X25519MLKEM768): on by default in Chrome since version 131, on most human traffic to Cloudflare, and on AWS KMS, ACM and Secrets Manager endpoints. Cloudflare targets 2029 for full post-quantum security, including authentication.Not their default for key exchange.Google · Cloudflare · Cloudflare roadmap · AWS
Research (individual authors)Bernstein (2026) estimates fewer breakable Ed25519+ML-DSA keys than ML-DSA-only keys and disputes arguments that hybrid signatures add no security. An individual Internet-Draft concludes that hybrid key establishment is preferable to standalone ML-KEM until a quantum computer breaks most classical keys. Neither has IETF consensus.Debated. Bernstein surveys and contests published arguments for going straight to PQC.Bernstein 2026 · draft-usama-tls-risks-of-mlkem

Hybrid KEMs

is the leading hybrid , combining Curve25519 ECDH with -768. Chrome enabled it by default in version 131 (November 2024, Google); by October 2025 it protected most human traffic to Cloudflare (Cloudflare); and AWS KMS, ACM and Secrets Manager endpoints support it (launched on non-FIPS endpoints in April 2025, AWS). AWS lists hybrid post-quantum key exchange as launched on FIPS and non-FIPS endpoints for AWS KMS, AWS Payment Cryptography and the AWS CloudHSM control plane. AWS pages differ on the edges, so check the endpoint and Region you use: the AWS KMS hybrid post-quantum TLS page says all endpoints including FIPS 140-3 validated endpoints in the Regions where AWS KMS is available, the AWS PQ readiness page lists FIPS and non-FIPS endpoints in commercial and US GovCloud Regions as launched, and the Infrastructure security in AWS KMS page says AWS KMS supports hybrid post-quantum TLS in all Regions except the China Regions and does not support it on AWS GovCloud (US) FIPS endpoints.

How X25519MLKEM768 Works

1.Key Generation: Generate both key pair and ML-KEM-768 key pair
2.Encapsulation: Sender performs X25519 ECDH + ML-KEM-768 encap → produces combined ciphertext
3.Decapsulation: Recipient performs X25519 ECDH + ML-KEM-768 decap → recovers combined shared secret
4.Key Derivation: Combined shared secret = KDF(both shared secrets, concatenated in the order the group defines) → session key

Even if ML-KEM-768 is broken in the future, X25519 still provides classical security. Even if X25519 is broken by a quantum computer, ML-KEM-768 provides quantum resistance.

Choosing a KEM Parameter Set (NIST SP 800-227)

FIPS 203 (§8) approves three ML-KEM parameter sets, each mapped to a NIST security category matching the strength of a given AES key size. Pick the set by threat model, not by default.

ML-KEMNIST CatClassical equiv.Typical deployment
ML-KEM-512Category 1AES-128Constrained/IoT, short-lived sessions
ML-KEM-768Category 3AES-192Default for TLS 1.3 & most internet traffic
ML-KEM-1024Category 5AES-256CNSA 2.0, high-assurance/federal systems

This module's advice: HNDL risk + data-retention window should drive the choice. If harvested ciphertext must remain secret past ~2035, prefer Cat 3 or Cat 5 over Cat 1.

Other Hybrid KEM Variants

SecP256r1MLKEM768 — P-256 + ML-KEM-768. Important for organizations requiring FIPS-approved classical curves (X25519 is not FIPS-approved).

SecP384r1MLKEM1024 — P-384 + ML-KEM-1024. Provides NIST Level 5 security for high-assurance environments.

Combiner Construction (SP 800-227 + SP 800-56C)

The high-level “KDF(both shared secrets)” above hides three real requirements that SP 800-227 defers to SP 800-56C-Rev2:

  • Concatenation order is fixed per group. In TLS 1.3 (RFC 10024 §4.3), X25519MLKEM768 concatenates ML-KEM_ss || X25519_ss, while SecP256r1MLKEM768 and SecP384r1MLKEM1024 concatenate ECDHE_ss || ML-KEM_ss; the result enters the TLS key schedule as its shared-secret input. SP 800-56C-Rev2 approves this when the first secret comes from a FIPS-approved scheme, so X25519MLKEM768 puts ML-KEM first (X25519 is not approved), while the NIST-curve groups put ECDHE first because certified P-256 and P-384 implementations are more widely available (RFC 10024 §5). SP 800-227 §4.6.2 (September 2025) relaxes the ordering rule: its key combiner is approved when at least one shared secret, in any position, comes from an approved scheme. Swapping order silently breaks interop. Order does not add security, but binding to a known order prevents cross-protocol confusion.
  • HKDF vs KMAC. SP 800-56C permits either an HMAC-based extractor (HKDF-SHA-256/384/512) or KMAC128/256. TLS 1.3 uses HKDF through its key schedule; KMAC is available for KEM-only protocols that want a single primitive.
  • Dual-PRF assumption. The combined secret stays safe as long as either input looks uniform to the attacker — so a future break of ML-KEM or of X25519 alone does not break the session key. This is why the classical half is kept during the transition.
  • Domain separation is strongly recommended. SP 800-227 §4.6.3 says a domain separator should be used to uniquely identify the composite scheme in use (for example the component schemes, order of composition, choice of parameter set, key combiner, KDF).

X-Wing and composite ML-KEM build the combiner into the KEM itself; TLS 1.3 relies on its own key schedule. The comparison below shows how they differ.

Combiners Compared, and X-Wing

Every hybrid KEM merges two shared secrets, but designs differ in what else goes into the final key. Hashing the exchanged values (ciphertexts, public keys) ties the key to this exchange; TLS gets the same effect from its transcript.

DesignWhat is combinedWhat ties it to the exchangeScopeSource
TLS 1.3 hybrid groupsThe two shared secrets, concatenated in the order the group defines, enter the TLS 1.3 key schedule (HKDF) as its shared-secret input.The handshake transcript, which includes both key shares.X25519MLKEM768, SecP256r1MLKEM768, SecP384r1MLKEM1024.RFC 10024
X-WingSHA3-256(ss_M ‖ ss_X ‖ ct_X ‖ pk_X ‖ XWingLabel)Hashes the X25519 ciphertext and public key. Leaves out the ML-KEM ciphertext and public key.ML-KEM-768 + X25519 only. A 32-byte private seed.X-Wing draft -11
Composite ML-KEM (LAMPS)SHA3-256(mlkemSS ‖ tradSS ‖ tradCT ‖ tradPK ‖ Label)The same shape as X-Wing: the traditional ciphertext and public key are hashed, with a label per algorithm.Twelve pairings with RSA-OAEP, ECDH (P-256, P-384, P-521, Brainpool), X25519 and X448, for X.509 and CMS.Composite ML-KEM -23
CFRG concrete hybrid KEMsThe generic CFRG framework with SHAKE256 as the PRG and SHA3-256 as the KDF, and a label per instance.The traditional ciphertext and public key, as in X-Wing.MLKEM768-P256, MLKEM768-X25519 (stated to be identical to X-Wing) and MLKEM1024-P384; used by HPKE.CFRG concrete -04
ETSI CatKDF / CasKDFCatKDF: secret = psk ‖ k1 ‖ k2, then one KDF (HMAC, KMAC or HKDF) with a label and context. CasKDF runs one KDF stage per component, chaining them.CatKDF puts all public keys and ciphertexts of both components into its context, which the composite ML-KEM draft calls the more conservative choice.Agreed in the EU certification mechanisms (ACM v2.0): TLS and IKE child SAs use the concatenate principle, the IKE SA uses the cascade.ETSI TS 103 744
What makes X-Wing different
  • It omits the ML-KEM ciphertext and key. The draft explains that X-Wing's security relies on the specifics of ML-KEM-768's Fujisaki-Okamoto transform, and warns that the combiner “cannot be assumed to be secure, when used with different KEMs”: it is “not known to be safe to leave out the post-quantum ciphertext from the combiner in the general case”.
  • Its security claim. If SHA3-256, SHA3-512 and SHAKE-256 are modelled as random oracles, X-Wing's IND-CCA security is bounded by that of ML-KEM-768 and the gap-CDH security of Curve25519.
  • Binding. The draft states that X-Wing binds the shared key to the public key and to the ciphertext (MAL-BIND-K-PK and MAL-BIND-K-CT), properties ML-KEM on its own does not achieve.
  • One fixed KEM with a seed key. X-Wing pairs only ML-KEM-768 with X25519. Its private key is a 32-byte seed that SHAKE256 expands into both component keys; the public key is 1,216 bytes, the ciphertext 1,120 bytes and the shared secret 32 bytes. That seed expansion is why the FIPS 140-3 guidance says X-Wing cannot yet be an approved hybrid service (see Implementation & Validation).
  • The same construction, standardized elsewhere. The CFRG's concrete MLKEM768-X25519 is stated to be identical to X-Wing. The composite ML-KEM draft calls X-Wing and its MLKEM768-X25519 entry “largely interchangeable”: the one difference is that X-Wing generates both private keys from the same seed, while composite ML-KEM assumes independently generated components, for example separately validated ones. An X-Wing implementation counts as compliant composite ML-KEM, but not the other way round.

— generate X25519 and ML-KEM-768 keys, run the encapsulation, then compare the HKDF, X-Wing and concatenation-only combiners on the same two secrets.

Hybrid KEMs in TLS 1.3

In TLS 1.3 (), hybrid KEMs integrate into the handshake via the key_share extension. The client and server exchange hybrid key material just as they would for classical ECDH — but with larger payloads:

Client→ClientHello supported_groups: [X25519MLKEM768, X25519]
key_share: ML-KEM-768 encapsulation key || X25519 public key (1,184 + 32 = 1,216 bytes)
Server→ServerHello selected_group: X25519MLKEM768
key_share: ML-KEM-768 ciphertext || X25519 public key (1,088 + 32 = 1,120 bytes)
Both→Shared secret = ML-KEM_ss || X25519_ss → 64 bytes (32 bytes each) (RFC 10024 §4.3)
Size Impact

Classical X25519 ClientHello key_share is 32 bytes. X25519MLKEM768 is 1,216 bytes (1,184 for ML-KEM plus 32 for X25519) — a 38× increase in the key share. Cloudflare notes that the ClientHello used to almost always fit within a single network packet; with a 1,216-byte key share it may now span more than one. RFC 6928 (Experimental) proposes an upper bound of 10 segments for TCP’s initial window (min (10*MSS, max (2*MSS, 14600))) in the first round-trip time, and a sender MAY start smaller.

Performance

The dominant cost is the larger key_share payload, not computation. Chrome reported about a 4% slowdown in TLS handshake time, larger on slow connections (as cited by Cloudflare, October 2025). AWS measured a 0.05% throughput drop for KMS calls that reuse connections, and 2.3% when every request opens a new handshake (April 2025).

For a full interactive TLS 1.3 handshake simulation with hybrid KEM negotiation, see the TLS Basics module.

Implementation and Validation (SP 800-227, FIPS 140-3)

A compliant ML-KEM deployment is more than “call the library.” SP 800-227 §3.2 covers the required input checks, and FIPS 203 specifies implicit rejection: if the re-encrypted ciphertext does not match the one received, decapsulation returns a key derived from the ciphertext and a secret value stored in the decapsulation key (Algorithm 18) instead of reporting an error. SP 800-227 §3.3 says KEM modules should be designed with appropriate countermeasures against side-channel attacks, including constant-time implementations and protecting memory from leakage.

ML-KEM Decapsulation and Implicit Rejection

ML-KEM is the key-encapsulation algorithm defined in FIPS 203. Alice uses Bob's public encapsulation key to produce a shared secret and a ciphertext. Bob uses his private decapsulation key and that ciphertext to recover the same secret. A protocol can then derive encryption and authentication keys from it.

Input checks can fail. For example, an ML-KEM-768 ciphertext must be 1,088 bytes long. A 1,087-byte ciphertext fails the length check, and decapsulation must not run. FIPS 203 also requires checks on the decapsulation key. Normal library argument and operational errors still apply.

An internal mismatch is different. For inputs that pass those checks, ML-KEM decrypts and re-encrypts internally, then compares the reconstructed ciphertext with the received one. If they differ, it selects a fallback shared secret derived from the secret rejection value z and the received ciphertext. The implementation returns that secret without exposing the mismatch as an error or through timing. This is implicit rejection.

What happens to the encrypted data? Suppose an attacker changes Alice's ciphertext without changing its length and triggers implicit rejection. Bob obtains a different secret. If the protocol derives an AES-GCM key from it, authentication of Alice's encrypted message fails, and the receiver rejects the message without releasing unauthenticated plaintext. Successful decapsulation alone does not confirm that both parties share a key; the protocol needs authentication or key confirmation.

Why hide the mismatch if authentication fails later? A specific decapsulation error would answer an attacker's repeated queries about the internal ciphertext check, creating a validity oracle. Implicit rejection conceals that answer. A later authentication failure only says that the message did not authenticate with the resulting key; a wrong tag can cause the same failure even after valid decapsulation. The protocol must preserve this distinction in its error handling and timing.

FIPS 203 §§6.3 and 7.3: internal decapsulation and input checks · SP 800-227: secure KEM use and key confirmation

Constant-Time Decapsulation

Correct results alone do not show that an implementation protects its secrets. Attackers may learn information by measuring execution time or observing memory access. Constant-time techniques help prevent these behaviours from revealing secret values; they do not mean every call takes exactly the same number of milliseconds. NIST recommends appropriate protections against these attacks. Their effectiveness depends on the implementation and the device running it.

Sources: SP 800-227 §3.3; RFC 7748 §5.1 for the X25519 component.

Randomness Requirements

ML-KEM encapsulation requires fresh, unpredictable random input (the 32-byte m). If an attacker knows that input, they can reproduce the encapsulation using the recipient’s public key and recover the same shared secret. FIPS 203 specifies the randomness requirements (§3.3), including use of an approved .

Side-Channel & Fault Resistance

Consider X25519 and ML-KEM combined to derive a session key. If a leak exposes only the X25519 shared secret, a suitable combiner can still protect the session key through the uncompromised ML-KEM secret. If the leak exposes both secrets or the final session key, that protection is lost. Assess what information the implementation leaks, including during combination and key handling.

For an HSM, TPM or other hardware device, check the manufacturer’s documentation to understand which protections are provided, which settings affect them, and which attacks they address. Do not assume every device has a switch to enable PQC protections.

Hybrid in a validated module (IG D.S)

Everything above is about how a hybrid KEM works. Whether a FIPS 140-3 validated module may offer it as an approved service is a separate question, answered by CMVP Implementation Guidance (IG) D.S, Key Encapsulation Mechanisms (last modified 9 April 2026). The IG calls an approved ML-KEM combined with one or more non-ML-KEM schemes “Scenario 2”. The non-ML-KEM part may be approved or non-approved (the IG's examples are X25519 and X448).

The five conditions: a module must implement, at a minimum, all of them

  1. An approved ML-KEM.
  2. One or more non-ML-KEM schemes, approved or non-approved.
  3. An approved key combiner: the shared secrets concatenated and fed into an SP 800-56C key-derivation function (HKDF counts), or combined by an SP 800-133 §6.3 method (concatenation, XOR, or key extraction using HMAC). The IG describes the SP 800-133 §6.3 route only for combining ML-KEM shared secrets with those of approved key-agreement schemes (IG D.F Scenario 1 or 2).
  4. The module does not offer arbitrary hybrid combinations. It enforces a pre-defined set of hybrid schemes, for example SecP256r1MLKEM768, X25519MLKEM768 and SecP384r1MLKEM1024.
  5. For each of those sets, the module controls at least how the shared secrets are sequenced, for example the concatenation order.

If any one is missing, Scenario 2 cannot be claimed, and any non-approved scheme the module offers must be restricted to the non-approved mode only. The IG's own examples: a software library with an approved ML-KEM and X25519 but no approved key combiner, or a library with the first three items that does not enforce a specific set of combinations or control the sequencing.

The non-approved half. A non-approved component such as X25519 is classified as a “non-approved but allowed algorithm with no security claimed” (IG 2.4.A): it is listed in the Security Policy, not on the certificate. It may be used only inside the approved hybrid, and the module itself must enforce that. It cannot be used for other purposes in the approved mode or offered as an independent approved service, such as a separate API call in a software library, and operator guidance in the Security Policy cannot stand in for that enforcement. The hybrid's claimed security strength comes from the ML-KEM, and the hybrid service is then treated as an approved security service.

Which real hybrids qualify. The IG says the TLS 1.3 hybrid key exchange (the IG cites an early version of the IETF draft that was later published as RFC 10024) may meet Scenario 2 with the SP 800-56C HKDF as the approved combiner, if one of SecP256r1MLKEM768, X25519MLKEM768 or SecP384r1MLKEM1024 is used, and notes that these specifications may still be in draft form. It also says the X-Wing hybrid KEM “cannot yet meet the requirements of Scenario 2” even though it uses an SP 800-56C one-step KDF as its combiner: at a minimum, SP 800-133 needs an update to approve X-Wing's seed generation, because seed expansion using SHAKE to generate the key pairs is not currently in SP 800-133. Separately, the CAVP prerequisites page (updated 22 September 2026) lists X-Wing keyGen and encapDecap tests. Both statements are published; CAVP itself says that acquiring a validation for each implemented algorithm does not by itself meet the FIPS 140 module requirements.

What the Security Policy shows, and what the lab tests

What the Security Policy shows. The module records the claim in the Security Function Implementation table as “KEM-Hybrid” or “KEM-Hybrid-Full”, with entry names such as “KEM-Hybrid X25519MLKEM768”. The non-approved half is listed separately with the caveat “no security claimed”, citing IG D.S Scenario 2 and IG 2.4.A. Per the CMVP's supplemental guidance (last updated 8 April 2026), “KEM-Hybrid” applies to a module that meets items 1 to 5 but not the extra conditions for “KEM-Hybrid-Full”, and it can still be an approved service, even for a software library offering the hybrid KEM to a calling application.

Self-tests. Every approved scheme in the hybrid must be self-tested individually under IG 10.3.A. For the ML-KEM part that means known-answer self-tests for each ML-KEM function the module implements (encapsulation, decapsulation and key generation), with the decapsulation tests covering both the implicit-rejection and the non-rejection path, on at least one of ML-KEM-512, ML-KEM-768 or ML-KEM-1024 as implemented in approved mode. The IG currently has no separate self-test for the approved key combiner or for the non-approved half, beyond self-testing any approved algorithms used within, such as KDFs.

Algorithm testing (CAVP). The IG lets the approved pieces be tested separately (the ML-KEM and, with a non-approved other half, the approved algorithms beneath it, such as KDFs). Its other option is end-to-end tests of the full hybrid scheme, “if available within CAVP”; it says it will be updated with the specific tests after they become available. No hybrid-KEM specification appears in the ACVP specification index (pages.nist.gov/ACVP) as of 23 October 2026.

Not to be confused with IG 1.B, “Hybrid Modules” (new 19 August 2026): that is about a module whose boundary combines a software or firmware component with a disjoint hardware component. It has nothing to do with hybrid key exchange.

In short: SP 800-227 sets out certain requirements and recommendations for KEM implementation and use in FIPS 140-validated modules, and gives guidance to vendors that choose to combine keying material securely.

HPKE — (RFC 9180)

Definition

HPKE is a public-key encryption scheme built from three swappable components — a , a KDF, and an AEAD — combined through a standard KeySchedule so any KEM/KDF/AEAD triple, plus one of four RFC 9180 modes (Base, PSK, Auth, AuthPSK), yields an interoperable construction. "Hybrid" in HPKE's own name refers to this KEM+KDF+AEAD composition, not to a PQC+classical pairing — that is a separate, additional meaning layered on top by draft-ietf-hpke-pq (below), and easy to conflate with the certificate-level hybrid formats covered earlier on this page.

Use cases

A KEM’s Encaps takes the recipient’s public key (plus a parameter set) and returns both a shared secret and the ciphertext; the sender does not hand it a key of its own choosing, whereas in key transport (for example RSA-OAEP) one party selects the key and transmits it. HPKE’s Encap is a randomized algorithm that generates an ephemeral shared secret and its encapsulation. Needing only the recipient’s public key is why it shows up wherever a protocol needs one-shot public-key encryption rather than an interactive handshake:

  • MLS uses HPKE for TreeKEM path updates (see the MLS Group Messaging module). Its post-quantum cipher suites (draft-ietf-mls-pq-ciphersuites-07) have been sent to the IESG with a request for publication.
  • TLS 1.3 Encrypted Client Hello (ECH, RFC 9849) uses it to encrypt the ClientHello.
  • JOSE and COSE extend it to wrap a JWT or CWT payload the same way. Status in November 2026: draft-ietf-jose-hpke-encrypt-22 is in the RFC Editor queue, the JOSE post-quantum suites (draft-ietf-jose-hpke-pq-pqt-02) are a working-group draft, and draft-ietf-cose-hpke-27 is with the IESG waiting for an Area Director go-ahead.

Variants

RFC 9180 itself registers only classical KEMs. The PQC angle comes from draft-ietf-hpke-pq — still an Internet-Draft, not yet an RFC (working-group draft, revision -05 of 6 July 2026) — which registers pure ML-KEM and PQ/T hybrid KEM IDs that plug into the same KeySchedule and Seal/Open.

RFC 9180 is being revised. The successor of RFC 9180, draft-ietf-hpke-hpke-06, removes the Auth and AuthPSK modes. The IETF Datatracker lists it in the RFC Editor queue (state on 8 November 2026). The draft is dated 29 October 2026. It adds a framework for single-stage KDFs, which is where the SHAKE and TurboSHAKE KDFs from draft-ietf-hpke-pq plug in. It also requires that applications not derive an AEAD key and nonce from an exported secret without fresh per-derivation input, and it requires implementations to support info values of at least 64 bytes and recommends support for at least 16,384 bytes.

DimensionRFC 9180 (ratified)draft-ietf-hpke-pq (draft)
KEMDHKEM(P-256/P-384/P-521/X25519/X448)ML-KEM-512/768/1024; MLKEM768-X25519, MLKEM768-P256, MLKEM1024-P384
KDFHKDF-SHA256/384/512+ SHAKE128/256, TurboSHAKE128/256
AEADAES-128/256-GCM, ChaCha20-Poly1305, Export-onlyunchanged

Independently of the KEM/KDF/AEAD choice, RFC 9180 defines four modes: Base (unauthenticated), PSK (pre-shared key), Auth (sender authenticated via its own KEM keypair), and AuthPSK (both). Auth and AuthPSK stay classical-only: draft-ietf-hpke-pq's ML-KEM and PQ/T hybrid KEM entries mark the Auth column "no", since ML-KEM has no AuthEncap/AuthDecap interface.

Auth and AuthPSK are RFC 9180 only. The successor draft does not specify them. It reserves their mode values (0x02 and 0x03) and does not deprecate them, so existing RFC 9180 deployments can keep using them as that RFC describes. The workshop below still offers all four modes so you can compare them, and labels the two RFC 9180-only modes.

The MLKEM768-X25519/P256 and MLKEM1024-P384 hybrid KEMs are not defined by draft-ietf-hpke-pq itself — it delegates to the CFRG's generic hybrid-KEM combiner (draft-irtf-cfrg-hybrid-kems), instantiated concretely in draft-irtf-cfrg-concrete-hybrid-kems: PQ-component-first concatenation, SHA3-256 as the combiner over concat(ss_PQ, ss_T, ct_T, ek_T, Label).

PKCS#11 v3.2 has no CKM_HPKE mechanism — checked against the canonical v3.2 header, no HPKE entry exists at all, and the v3.3 working draft has none either (the standards committee has only an open request). A real HSM deployment composes it from mechanisms the spec does define: CKM_ECDH1_DERIVE / CKM_ML_KEM for the KEM leg, plus HMAC signing for LabeledExtract/LabeledExpand, and AES-GCM / ChaCha20-Poly1305 for Seal/Open. — every KEM/KDF/AEAD/mode combination above is selectable, and the classical DHKEM path is cross-checked byte-for-byte against RFC 9180 Appendix A.

Hybrid Signatures

Signing with both a classical and a PQC algorithm at the same time. The key design question: can either component be stripped and verified alone?

RFC 9955: Hybrid Signature Spectrums (Informational, July 2026)

Defines the taxonomy of hybrid signature constructions and a non-separability spectrum. Basis for evaluating all hybrid signature proposals including Silithium.

rfc-editor.org ↗
ConstructionHow it worksNon-separabilityBackwards compat
Concatenationsig₁ ‖ sig₂ None / WNSYes
Nestingsign_ML(msg ‖ sig_EC) WNSYes
Fused (Silithium; the workshop uses a local model)shared challenge μ (workshop model: H(R ‖ pk_ec ‖ pk_ml ‖ msg)) SNS (targeted by the paper; not shown by the workshop)No

Non-Separability Spectrum

None
Either sig strips cleanly
WNS
Swapping invalidates outer
SNS
Neither verifies alone
SNS + Sim
Both must verify simultaneously

Silithium — Compact Fused Hybrid Signature

Devevey, Guerreau, Roméas · ePrint 2025/2059 · November 2025

ePrint

Silithium adapts the Fiat-Shamir transform to produce a fused hybrid signature from EC-Schnorr (classical, secp256k1) and ML-DSA-65 (FIPS 204). Both components share a single challenge μ derived from both public keys, the EC commitment R, and the message. Neither component verifies on its own with the standard challenge. The paper targets Strong Non-Separability (SNS) and reports no separability issues for its construction; this page demonstrates the idea and does not prove the property.

Protocol (Sign)
1
EC commitment: k ← fresh random scalar (local model), R = k·G (secp256k1 base point)
2
Shared challenge: Local model: μ = H(R ‖ pk_ec ‖ pk_ml ‖ msg), passed to the provider as ML-DSA external μ
3
EC response: s = (k + e·sk_ec) mod n where e = μ[:32] as bigint
4
ML-DSA response: z = ML-DSA.Sign_internal(sk_ml, μ) (FIPS 204 §6.2, Algorithm 7, external μ)
5
Output: σ = R ‖ s ‖ z (same total size as concatenation — savings require full w1 binding)
Hybrid EU-CMA
Paper's claim for the full scheme: a forgery requires breaking both EC-Schnorr and ML-DSA (this simplified demo is not covered by that proof)
SNS targeted
The paper's goal. This page's local model only shows that the standard raw-message checks fail; that observation is not a proof.
No backwards compat
Requires a verifier that understands the fused scheme

Attack Model: What Non-Separability Prevents

Separability Attack
Adversary strips the PQC component and presents only the classical signature. Verifiers that accept classical-only sigs are fooled. What SNS is meant to prevent.
Recombination Attack
Adversary combines a legitimate EC sig with a different ML-DSA sig (or vice versa) to forge a hybrid signature. The local model ties both parts to the same R and both public keys; the workshop tests one replacement case only.
Cross-Protocol Attack
A valid signature in one hybrid scheme is replayed in another. Prevented by including both pk_ec and pk_ml in the challenge, tying the sig to both key pairs.

IG D.S covers key encapsulation only. The IG (last updated 19 August 2026) has no section on composite or hybrid signatures. It does name one hybrid key-establishment scheme that uses ML-KEM (IG D.S, “Hybrid scheme that uses an approved ML-KEM”); a hybrid module (IG 1.B) is a different thing, about how a module is built, not about combining algorithms. What the IG does say applies to each algorithm on its own: a module that implements ML-DSA signing or verification needs the self-tests of IG 10.3.A. Whether a composite signature can be offered as an approved service is not answered in the published guidance; ask the lab.

Certificate Format Approaches

Seven X.509 certificate formats make up the workshop's main comparison: five carry signature keys and two carry KEM (encryption) keys. One rule underlies all of them: the subject public key and the certificate signature are independent. The subject key is what the certificate is about; the signature comes from the issuer's key.

PQC-only — no legacy fallback

Pure PQC (ML-DSA)

Published
Certificate {
algorithm: id-ml-dsa-65
OID: 2.16.840.1.101.3.4.3.18
publicKey: ML-DSA-65-key
signature: ML-DSA-65-sig
}

Standard single-algorithm X.509 using a lattice-based PQC signature. ML-DSA OIDs are standardized in RFC 9881. Quantum-safe but requires all verifiers to support ML-DSA.

Pure PQC (SLH-DSA)

Published
Certificate {
algorithm: id-slh-dsa-sha2-128s
OID: 2.16.840.1.101.3.4.3.20
publicKey: SLH-DSA-128s-key
signature: SLH-DSA-128s-sig (7856 B)
}

Hash-based signature (SPHINCS+) with no lattice assumptions. OIDs in RFC 9909. does not recommend systematic hybridisation for hash-based signatures such as SLH-DSA, XMSS and LMS, because it considers their underlying problem well studied (its 2022 position paper also notes their limited range of application). Larger signatures (7,856 B, about 7.9 KB, for SLH-DSA-SHA2-128s per FIPS 205) are the trade-off.

Composite (Dual-Algorithm)

RFC 10073
Certificate {
algorithm: id-MLDSA65-ECDSA-P256-SHA512
OID: 1.3.6.1.5.5.7.6.45
publicKey: ML-DSA-65-key || ECDSA-key
signature: ML-DSA-65-sig || ECDSA-sig
}

A single composite OID identifies the algorithm pair. Both signatures must verify. Under RFC 10073 §9.2.3, the traditional component can verify over the composite’s prefixed input. The ML-DSA component also requires the composite context. X.509 verification rejects a stripped component because the signed data still names the composite algorithm. The validator must support that composite OID.

Hybrid with legacy fallback

Alt-Sig / Catalyst

ITU-T X.509
Certificate {
algorithm: ecdsa-with-SHA256
extensions {
AltPubKey (2.5.29.72): ML-DSA-65
AltSigAlg (2.5.29.73): ML-DSA-65
AltSigValue (2.5.29.74): ML-DSA-sig
}
}

A single classical cert with PQC key and signature in X.509 extensions. Legacy verifiers ignore the extensions; PQC-aware verifiers can check the alternative signature too, and their policy decides whether one or both must verify. Defined in ITU-T X.509 (2019) §7.2.2 and §9.8.

Related Certs (RFC 9763)

Published
Existing Cert A (Classical) {
algorithm: ecdsa-with-SHA256
(unchanged)
}
New Cert B (PQC) {
algorithm: ML-DSA-65
ext: RelatedCertificate =
hash(final DER of Cert A)
}

A new certificate carries a RelatedCertificate extension holding the hash of an existing one. The link is one-way: Cert A is never modified. The CA issues Cert B only after the requester proves it holds Cert A's key. Legacy systems use the classical cert; a protocol may ask for either certificate or both — RFC 9763 does not require both.

KEM certificates — the subject key encrypts, the issuer signs

ML-KEM subject key ── performs encapsulation / decapsulation
ML-DSA issuer key ── signs the X.509 certificate

Pure PQC KEM (ML-KEM)

Published

RFC 9935 defines ML-KEM public keys in X.509. An ML-KEM key cannot sign, so its certificate is always issued by a separate signing-capable CA (ML-DSA, ECDSA, RSA, …). If keyUsage is present, keyEncipherment must be the only bit. Because the CA cannot check a self-signature from a KEM key, proof of possession uses a KEM-based exchange instead. The key is then used, for example, in CMS KEMRecipientInfo (RFC 9629, ML-KEM in RFC 9936).

Composite KEM

RFC Editor queue

draft-ietf-lamps-pq-composite-kem-23 puts an ML-KEM key and a classical KEM key (X25519, ECDH, RSA) under one OID, ML-KEM component first. Encapsulation runs both and combines the shared secrets. Approved by the IESG and awaiting publication as an RFC; not backward compatible.

Advanced examples

Certificate Discovery

Active draft

draft-ietf-lamps-certdiscovery-06 lets a certificate tell PQC-capable relying parties where to fetch a related certificate, through a subjectInfoAccess entry that can also name the related certificate's algorithms and carry its hash. It does not say why the two are related (earlier drafts had an “intent” field; -06 removed it). A fetched certificate must still pass full path validation. The OIDs are not assigned yet, so the workshop's encoding is illustrative.

Unsigned certificate

Published

RFC 9925 defines id-alg-unsigned for a certificate with no signature at all — a container for a key that is trusted by other means, such as a trust anchor. It is never valid in a certification path.

Historical design

Chameleon

Expired draft

One certificate with a DeltaCertificateDescriptor extension encoding the differences needed to rebuild a paired certificate. The proposal, draft-bonnell-lamps-chameleon-certs-07, is an expired individual draft that the LAMPS working group never adopted. It is shown for study only and is not recommended for new deployments.

Deployment guidance: This module's advice: Pure PQC (ML-DSA) is ready today (RFC 9881; the workshop is tested with OpenSSL 3.6.3). Pure SLH-DSA is also ready (RFC 9909). Composite requires both parties to support the draft spec — best suited to closed PKI. Alt-Sig works with legacy verifiers via extensions. Related Certs (RFC 9763) keeps the classical certificate fully usable on its own. Watchlist: FN-DSA (planned FIPS 206, not yet final) has an IETF certificate draft whose OIDs are still TBD; composite KEM will enable hybrid key encapsulation in S/MIME once published.

Hybrid in Crypto APIs and HSMs

An application gets a hybrid in one of two ways. It can compose it: call a classical API and a pure-PQC API separately, then combine the results itself. Or the library or token offers the hybrid as one algorithm: a TLS named group, a composite KEM such as X-Wing or composite ML-KEM, or a composite signature. Today most libraries offer hybrid key exchange only inside TLS, so any hybrid outside TLS is usually composed by the application.

How applications compose a hybrid today

  • Key establishment in Java: an X25519 KeyAgreement, an ML-KEM KEM and the KDF API's HKDF, fed both secrets in an order the application fixes. JEP 527 notes the platform “already contains the building blocks”, but adds hybrid itself only for TLS 1.3; Bouncy Castle's guidance is that “the order is part of the protocol and must be fixed by both parties in advance”. JEP 527 · JEP 510 · BC specifications
  • Key establishment on an HSM (PKCS#11 v3.2): CKM_ECDH1_DERIVE for the classical secret, C_EncapsulateKey / C_DecapsulateKey with CKM_ML_KEM for the post-quantum one, then the application either reads both out and combines them, or chains CKM_CONCATENATE_BASE_AND_KEY, CKM_CONCATENATE_BASE_AND_DATA and CKM_SHA3_256_KEY_DERIVATION inside the token, as this module's HPKE step does. PKCS#11 v3.2
  • Signatures: two separate signatures, one classical and one ML-DSA, each with its own key, and a verification policy in the application that requires both. Microsoft's 2025 guidance still recommended this kind of combination with existing algorithms before its composites shipped. Microsoft, 2025

Limits of composing it yourself, especially with an HSM

  • The application owns the combiner, and can get it wrong

    Order, label, KDF and what gets bound are all the caller’s choice. A naive "concatenation only" key is not secure against active adversaries, and different protocols use different combiners, so two applications built on the same primitives can disagree. Microsoft gives this as the reason for native composites: they "abstract away the complexity of securely combining multiple algorithms, reducing the risk of incorrect integrations".

    ETSI TR 103 966 · Microsoft, 2026
  • Secrets leave the HSM unless every template is right

    In PKCS#11 v3.2, C_EncapsulateKey and C_DecapsulateKey create a secret-key object whose CKA_EXTRACTABLE defaults to true unless the template says otherwise, and whose CKA_ALWAYS_SENSITIVE and CKA_NEVER_EXTRACTABLE are set to false. Combining the two secrets in application memory means reading them out of the token, and anything that can read that memory can rebuild the session key. Combining inside the token is possible, but only as a chain the application drives.

    PKCS#11 v3.2
  • The token does not know it is computing a hybrid

    A v3.2 token can concatenate two secret keys (CKM_CONCATENATE_BASE_AND_KEY), append public data (CKM_CONCATENATE_BASE_AND_DATA) and hash the result (CKM_SHA3_256_KEY_DERIVATION, or CKM_HKDF_DERIVE). Each step is a separate C_DeriveKey with its own template, and the token checks none of the order, label or hash choice. For FIPS 140-3, a hybrid KEM is an approved service only when the module itself enforces a fixed set of hybrid schemes and the order of the secrets (IG D.S Scenario 2).

    PKCS#11 v3.2 · FIPS 140-3 IG
  • Two signature keys that nothing ties together

    C_SignInit takes one key handle, so an application-composed hybrid signature is two C_Sign calls on two key objects. PKCS#11 v3.2 has no mechanism that binds the keys or requires both signatures; the verifier’s policy must insist on both, or one signature can be stripped. The two keys also need their own lifecycle, attributes and backup, kept in step by the application.

    PKCS#11 v3.2 · ETSI TR 103 966

What a single hybrid mechanism changes: PKCS#11 v3.3

The PKCS#11 v3.3 working draft adds key types CKK_COMP_KEM and CKK_COMP_SIG and five mechanisms: CKM_COMP_KEM_KEY_PAIR_GEN and CKM_COMP_KEM (run through C_EncapsulateKey / C_DecapsulateKey, returning the combined secret as a key handle), and CKM_COMP_SIG_KEY_PAIR_GEN, CKM_COMP_SIG and CKM_COMP_SIG_EXTERNAL_HASH (one C_Sign call for one composite signature from one key object holding both halves). The combiner then runs inside the token, as the composite ML-KEM and ML-DSA drafts define it. The OASIS PKCS 11 TC approved these identifiers in October 2026, together with identifiers for KMAC and the SHAKE functions; no v3.3 specification has been published. The draft has no HPKE mechanism, only an open request for one. For three combinations (MLKEM768-X25519, which is X-Wing, MLKEM768-P256 and MLKEM1024-P384) the composite KEM computes the same secret as the matching HPKE hybrid KEM, so it can do the key-agreement step of HPKE. TC header · composite KEM text · composite signature text

Try both mechanisms in the Hub's PKCS#11 engine: composite ML-KEM and composite ML-DSA in the PKCS#11 workshop's v3.3 tab. A single mechanism is not automatically an approved FIPS 140-3 service: the published guidance has no section on composite signatures (see Implementation & Validation).

Which libraries and APIs offer what (1 November 2026)

Library / APIPure PQCHybrid TLS 1.3 groupsHybrid KEM as one algorithmComposite signature as one algorithmSources
PKCS#11 v3.2 (OASIS Standard)ML-KEM, ML-DSA mechanismsn/aNot offered: composed from separate mechanismsNot offeredv3.2
PKCS#11 v3.3 (working draft)As v3.2n/aCKM_COMP_KEM (12 parameter sets; draft)CKM_COMP_SIG, CKM_COMP_SIG_EXTERNAL_HASH (18 sets; draft)TC header · KEM text · sig text
Java JDK 27 (and 25.0.5)ML-KEM, ML-DSA since JDK 24X25519MLKEM768 default; SecP256r1MLKEM768, SecP384r1MLKEM1024 opt-inNot offered: hybrid only inside TLS, by designNot offeredJEP 527 · JDK 25.0.5 · JEP 496 · JEP 497
Bouncy Castle Java 1.86ML-KEM, ML-DSA, SLH-DSA and moreAll three groups in the BCJSSE default listComposite ML-KEM (experimental); X-Wing in the lightweight APIComposite ML-DSA (18 combinations)release notes · 1.86 · specifications
Windows CNG and SChannel (Windows 11 24H2/25H2, Server 2025)ML-KEM, ML-DSA (GA November 2025)All three groups, not enabled by defaultComposite-ML-KEM (GA July 2026)Composite-ML-DSA, four ECDSA sets (GA July 2026)PQC APIs GA · composites · TLS FAQ · CNG identifiers
.NET 10 / .NET 11MLKem, MLDsa (.NET 10)From the operating systemCompositeMLKem (.NET 11, experimental)CompositeMLDsa (experimental; native CNG on Windows in .NET 11).NET PQC · .NET 11 · native composite
Microsoft SymCryptML-KEM, ML-DSAn/aComposite ML-KEM (v103.11.0)Composite ML-DSA (v103.12.0)changelog
OpenSSL 3.5 LTS to 4.1ML-KEM, ML-DSA, SLH-DSAX25519MLKEM768 default; 4.x also defaults SecP256r1MLKEM768; SecP384r1MLKEM1024 availableNot offeredNot offered4.1.0 · 3.5 groups · 3.6
BoringSSLML-KEM, ML-DSA, SLH-DSAX25519MLKEM768 defaultX-Wing (xwing.h, and as an HPKE KEM)Not offeredxwing.h
AWS-LC 5.12ML-KEM, ML-DSAAll three groups, in the default listNot offeredNot offeredv5.12.0
Go 1.24 to 1.27crypto/mlkem (1.24), crypto/mldsa (1.27)X25519MLKEM768 default since 1.24; all three since 1.26crypto/hpke: MLKEM768X25519 (X-Wing), MLKEM768P256, MLKEM1024P384 (1.26)Not offered1.24 · 1.26 · 1.27 · crypto/hpke
rustls 0.23 (aws-lc-rs provider)ML-KEM groups; ML-DSA certificatesX25519MLKEM768 preferred by default; SecP256r1MLKEM768 opt-inNot offeredNot offered0.23.46 · 0.23.27
Mozilla NSS / FirefoxML-KEM, ML-DSAX25519 + ML-KEM-768 on by default in Firefox since 132; NIST-curve hybrids in NSS 3.118Not offeredNot offeredNSS 3.132 · NSS 3.118 · Firefox 132
wolfSSL 5.9.4ML-KEM, ML-DSAAll three groups with hybrids built in (on by default); the default key share is a standardized hybrid when ML-KEM and X25519 or P-256 are built inNot offeredNot offered5.9.4
liboqs 0.16 / oqs-provider 0.12ML-KEM, ML-DSA, SLH-DSA and moreHybrid groups offered; on OpenSSL 3.5+ OpenSSL’s own are usedNot offeredNot offered: the experimental composite integration was removedliboqs 0.16.0 · oqs-provider 0.12.0

“Not offered” means the project's release notes and documentation list no such algorithm as of 1 November 2026. The TLS groups are those of RFC 10024: X25519MLKEM768, SecP256r1MLKEM768 and SecP384r1MLKEM1024. The Migrate catalog shows the same information per product.

Related Resources

Generate hybrid keys, run KEM operations, and inspect .

Check off all sections and mark this reading done.

In the Industry Landscape

Check your understanding

28 questions on Hybrid Cryptography, each with its answer and the reason.

Take the quiz

Next step

Practice it: Hybrid Certificates

Hybrid Certificates is the hands-on version of this module: the same ideas, run in your browser.

Production Tool Capabilities & Verification Checklist

Capabilities to require in your production tools:

  • Negotiate an RFC 10024 hybrid key agreement group in TLS 1.3 and record the group selected by the server.
  • Verify the peer name, certificate path, and actual CertificateVerify signature against the authentication policy. RFC 10073 defines composite ML-DSA in X.509. Composite ML-DSA in TLS 1.3 has a separate Internet-Draft.
  • Enforce the expected group, signature and trust policy. Exercise rejection of a classical-only peer when hybrid key agreement is required, a peer with a disallowed signature, and a certificate with an unexpected name or trust anchor.

How to evaluate in your environment

Use the exact client, server, provider and trust configuration planned for rollout. Record a classical baseline and test each intended certificate choice on both peers. Capture the selected key agreement separately from peer verification and CertificateVerify.

Production evidence to collect

Build and configuration identifiers, negotiated group, peer certificate and trust anchor identifiers, the verified CertificateVerify scheme, and results of expected rejections.

Completing this Hub module and its browser-based sample exercises teaches the technical and governance decision criteria for your role; it does not test, validate, or certify your deployed production systems.

Record evidence for your environment

Identify the tested build and configuration. Use reference IDs for observations. For a proposed exclusion, record the system boundary or rule and why it excludes this build.

Negotiate an RFC 10024 hybrid key agreement group in TLS 1.3 and record the group selected by the server.
Verify the peer name, certificate path, and actual CertificateVerify signature against the authentication policy. RFC 10073 defines composite ML-DSA in X.509. Composite ML-DSA in TLS 1.3 has a separate Internet-Draft.
Enforce the expected group, signature and trust policy. Exercise rejection of a classical-only peer when hybrid key agreement is required, a peer with a disallowed signature, and a certificate with an unexpected name or trust anchor.

Evidence missing. 0 of 3 requirements have a reference or scope rationale.

Learning module content can be inaccurate. Please double-check its information. Report inaccuracies in PQC Today GitHub Discussions.