N-PAMP cryptographic provider matrix (ML-KEM + classical ECDH)¶
Dated 2026-08-21; every version and date below was confirmed against a primary source (the vendor's own release page, package registry, or standards document) on that date. Provider ecosystems move quickly — re-confirm any row before it gates a build decision.
N-PAMP's two hybrid key-exchange groups each combine a post-quantum ML-KEM component with a classical elliptic-curve Diffie–Hellman component:
- X25519MLKEM768 (code point
0x11ec; Standard/High profiles) — ML-KEM-768 + X25519. - SecP384r1MLKEM1024 (code point
0x11ed; High/Sovereign profiles) — ML-KEM-1024 + P-384.
The hybrid-KEM combiner adapter is provider-agnostic: it consumes the two raw component shared secrets a language's own ML-KEM and ECDH providers produce, and concatenates them in the per-group order fixed by the specification. This document records which provider each reference SDK language should use to produce those two component secrets. It is implementation guidance, not a wire specification — the wire bytes are fixed by the spec regardless of which library computes them.
Ecosystem baseline¶
- OpenSSL 3.5.0 (released 2025-04-08, an LTS line supported to 2030-04-08) ships native
ML-KEM (FIPS 203), ML-DSA (FIPS 204), and SLH-DSA (FIPS 205) in its default provider, and
changed its default TLS key shares to offer
X25519MLKEM768. From 3.5 onward, standard ML-KEM needs no add-on provider. OpenSSL 3.6.0 (2025-10-01) carries the same PQC baseline forward. - liboqs 0.16.0 (2026-07-09) implements ML-KEM via the PQCP
mlkem-nativeproject. oqs-provider 0.11.0 (2025-12-24) now defers to OpenSSL's own ML-KEM/ML-DSA on hosts running OpenSSL ≥ 3.5, rather than shipping a competing implementation. liboqs remains useful as an independent cross-check source and for hosts pinned below OpenSSL 3.5. - FIPS boundary caveat: "ML-KEM is present in the OpenSSL 3.5 library" is a separate claim from "ML-KEM is in a FIPS-140-3-validated OpenSSL module" — the two are validated on different CMVP timelines. Whether a deployment needs library-level or validated-module-level PQC is a per-deployment decision, tracked in the crypto-agility runbook, not here.
- Profile alignment: the SecP384r1MLKEM1024 group (ML-KEM-1024 + P-384) is the one aligned
with an ML-KEM-1024-only posture; X25519MLKEM768 (ML-KEM-768) is the broader-interop group.
This mirrors
RFC 10024's (formerlydraft-ietf-tls-ecdhe-mlkem) own two-suite structure.
Per-language provider matrix¶
| Language | Recommended ML-KEM provider | Version / date | Classical half (X25519 / P-384) | Maturity |
|---|---|---|---|---|
| Go | stdlib crypto/mlkem |
Go 1.24 (2025-02-11) | stdlib crypto/ecdh |
Native — the reference path; no external dependency |
| Rust | RustCrypto ml-kem (pure Rust) or aws-lc-rs |
ml-kem 0.3.2 (2026-05-10) |
aws-lc-rs (bundled) or p384 + x25519-dalek |
Solid; ml-kem hazmat exposes deterministic encaps for KATs |
| Python | pyca cryptography (native, Rust-backed) |
48.0.0 (2026-05-04) | cryptography hazmat EC |
Solid; ecosystem-standard library, zero new supply-chain surface |
| JavaScript / TypeScript (Node) | built-in node:crypto |
Node 24+ (confirmed on v26.7.0 docs) | node:crypto (X25519, secp384r1) |
Native; needs Node's linked OpenSSL/BoringSSL ≥ 3.5. Browser/edge fallback: @noble/post-quantum |
| Java | Bouncy Castle via javax.crypto.KEM (Java 17+) |
BC 1.84 (2026-05-12); FIPS path BC-FJA 2.1.0 | BC X25519 / secp384r1 | Solid; de-facto JVM PQC standard (no OpenJDK built-in yet) |
| C# / .NET | System.Security.Cryptography.MLKem (Linux/Windows) |
.NET 10 (2025-11) | ECDiffieHellman (P-384); BC for X25519 |
Native on Linux/Windows; macOS gap → BouncyCastle.Cryptography 2.7.0 (self-labeled experimental), native macOS deferred to .NET 11 |
| Ruby | no mature native binding yet | tracked at ruby/openssl#894 (open) + bugs.ruby-lang.org #22068 (open) |
OpenSSL::PKey (X25519, P-384) |
Gap — the ruby/openssl gem's PQC path (provider-pkey passthrough) will inherit ML-KEM once the tracking issue lands against OpenSSL ≥ 3.5; not shipped as of 2026-08-21. See notes |
| PHP | paragonie/ext-pqcrypto (Rust-backed extension) |
0.1.0 (2026-04-07) | core openssl ext (X25519, P-384) |
Workable; core PHP openssl does not expose ML-KEM (absent from the PHP 8.6 RFC roster) → a non-core extension is required |
| Swift | SwiftKyber (pure Swift, cross-platform) | 3.5.0 (2026-04-21); deterministic DeriveKeyPair since 3.4.0 |
CryptoKit Curve25519 / P384 |
Solid; NIST-ACVP-sourced KAT suite (non-circular). Preferred over Apple CryptoKit (Apple-platform-only) |
The Go reference provider path (already wired)¶
The Go reference implementation already sources both component secrets of X25519MLKEM768 from the standard library, with no external cryptographic dependency (the open reference does not implement SecP384r1MLKEM1024):
impl/go/kem.go— X25519MLKEM768:crypto/mlkem(GenerateKey768/Encapsulate/Decapsulate) +crypto/ecdh(X25519).
impl/go/hybridkem/provider_path_test.go demonstrates the end-to-end path in isolation: it
runs a live X25519MLKEM768 exchange using those stdlib providers,
then feeds the resulting raw component secrets into the standalone combiner adapter
(hybridkem.Combine) and asserts the adapter reproduces the reference key-schedule input
byte-for-byte, in the ML-KEM-first order of 0x11ec. This is the concrete link between the provider and the combiner specification.
Notes, gaps, and checks still to make¶
- Ruby is the one language without a mature native ML-KEM path today. The
ruby/opensslgem (latest 4.0.2, 2026-05-13) tracks PQC support upstream atruby/openssl#894, and the broader Ruby-stdlib effort isbugs.ruby-lang.orgFeature #22068 — both open and in active, dated development as of 2026-08-21. There is no OQS-official liboqs-ruby wrapper and no dedicated ML-KEM gem was located. The Ruby SDK's live ML-KEM path is therefore a prerequisite to re-confirm, not a settled dependency: re-check whether the tracking issue has landed before the Ruby stack is built, since the upstream work is moving. - PHP has no core-
opensslML-KEM (confirmed absent from the PHP 8.6 RFC roster); it depends on a non-core extension (paragonie/ext-pqcrypto, or a liboqs FFI binding). This is a deployment-friction note, not a blocker. - Deterministic / seed-based keygen (needed to drive standards-anchored known-answer
vectors through the real code path) is confirmed present for Go, Rust, Node, and Swift. For
Python
cryptography, Java Bouncy Castle, .NET, and PHPext-pqcrypto, the exact deterministic-keygen call surface was not confirmed in this pass — verify it before wiring each language's KAT harness. - CNSA 2.0 posture (ML-KEM-1024-only, milestone dates) is recorded from secondary summaries in this pass; open the NSA CNSA 2.0 FAQ directly before any CNSA-2.0 conformance claim.
Full research provenance (primary sources, methodology, and the complete limitations list) is retained in the internal Part-2 build notes.