Post-Quantum Cryptography: Threat, Standards and Timelines
Contents
This is part 1 of a two-part series. It covers post-quantum cryptography (PQC) in general: the threat, the organisations that set the rules, the concepts and the algorithms. Part 2, Post-Quantum Cryptography in IPsec, applies all of this to IKEv2 and IPsec VPNs.
1. Why PQC, and how urgent is it?
1.1 The threat
All public-key cryptography in IPsec today rests on two mathematical problems: integer factorisation (RSA) and discrete logarithms in finite fields or on elliptic curves (MODP Diffie-Hellman, ECDH, ECDSA, EdDSA). Shor’s algorithm solves both efficiently on a large, fault-tolerant quantum computer, usually called a cryptographically relevant quantum computer (CRQC). Once a CRQC exists, every IKEv2 key exchange and signature that uses these algorithms is broken.
Symmetric cryptography is affected far less. Grover’s algorithm gives at most a quadratic speed-up for brute-force key search, and it parallelises poorly. AES, SHA-2 and HMAC therefore stay usable. Larger parameters (AES-256, SHA-384) give a comfortable margin.
1.2 Security strength: classical vs. quantum
The difference is easiest to see in bits of security. Against a classical computer, RSA-3072, P-256 and AES-128 all offer about 128 bits. Against a CRQC, RSA and elliptic curves drop to effectively zero, while AES keeps half of its nominal strength at worst.
Bits of security (higher is stronger). The solid line is the 128-bit classical baseline. The dashed line is AES-128 under Grover, the floor NIST accepts for Category 1. Hover a bar for details.
Show data as table
| Algorithm | Classical | vs. CRQC | Why |
|---|---|---|---|
| RSA-2048 | 112 | broken | Factoring: broken by Shor |
| RSA-3072 | 128 | broken | Factoring: broken by Shor |
| DH MODP-3072 | 128 | broken | Discrete log: broken by Shor |
| ECDH/ECDSA P-256 | 128 | broken | Elliptic-curve discrete log: broken by Shor |
| X25519 / Ed25519 | 128 | broken | Elliptic-curve discrete log: broken by Shor |
| ECDH/ECDSA P-384 | 192 | broken | Elliptic-curve discrete log: broken by Shor |
| AES-128 | 128 | ≈64 | Grover: nominal halving is an upper bound. Grover parallelises poorly; a 2026 analysis (Valsorda) puts a realistic attack at about 2^104 operations. NIST still uses AES-128 as the Category 1 yardstick |
| AES-192 | 192 | ≈96 | Grover: nominal halving |
| AES-256 | 256 | ≈128 | Grover: nominal halving; comfortable margin |
| ML-KEM-512 | ≥128 | ≥64 | Category 1: at least as hard as AES-128 key search, classical and quantum |
| ML-KEM-768 | ≥192 | ≥96 | Category 3: at least as hard as AES-192 key search |
| ML-KEM-1024 | ≥256 | ≥128 | Category 5: at least as hard as AES-256 key search |
| ML-DSA-65 | ≥192 | ≥96 | Category 3: at least as hard as AES-192 key search |
| ML-DSA-87 | ≥256 | ≥128 | Category 5: at least as hard as AES-256 key search |
| SLH-DSA-128s/f | ≥128 | ≥64 | Category 1: at least as hard as AES-128 key search |
| SLH-DSA-256s/f | ≥256 | ≥128 | Category 5: at least as hard as AES-256 key search |
Post-quantum algorithms are not in this chart, because they are not rated in bits but in NIST security categories. Each category is defined as “at least as hard to break as” a reference symmetric attack, under classical and quantum attack (section 4).
1.3 How far away is a CRQC?
Nobody knows, but the estimated resources needed keep falling, and faster than expected:
- 2019: Gidney and Ekerå estimated that RSA-2048 could be factored in about eight hours with some 20 million noisy physical qubits.
- May 2025: Craig Gidney (Google Quantum AI) cut that to fewer than one million noisy qubits running for under a week, under the same hardware assumptions.
- March 2026: a Caltech and Oratomic preprint (arXiv:2603.28627) proposed high-rate error-correcting codes for neutral-atom machines. Today about 1,000 physical qubits are often needed per logical qubit; this approach could need as few as five. That leads to estimates of roughly 10,000–26,000 physical qubits to break 256-bit elliptic-curve cryptography in about ten days, and similar numbers for RSA-2048 with runtimes of months to years. The work is theoretical, not yet peer-reviewed, and assumes machines that still have to be built. Neutral-atom arrays with more than 6,000 atoms have been demonstrated.
- Spring 2026: Google researchers described a more efficient version of Shor’s algorithm against elliptic curves. They published it as a zero-knowledge proof, showing feasibility without revealing the method.
Logarithmic scale; each point is a published resource estimate, not a machine that exists. Assumptions differ (superconducting vs. neutral-atom hardware, runtime). Hover a point for details.
Show data as table
| Year | Estimate | Source and assumptions |
|---|---|---|
| 2012 | ≈1 billion | 2012 estimate cited by Google Quantum AI: about a billion physical qubits |
| 2019 | ≈20 million | Gidney & Ekerå (2019): about 20 million noisy qubits, ~8 hours (superconducting assumptions) |
| 2025 | <1 million | Gidney (May 2025): fewer than one million noisy qubits, under a week (same hardware assumptions as 2019) |
| 2026 | ≈11k–26k | Caltech & Oratomic preprint (Mar 2026, arXiv:2603.28627): ~11,000–26,000 qubits on neutral-atom hardware with high-rate error-correcting codes, months to years of runtime. Theoretical, not peer-reviewed. |
Real demonstrations remain tiny. In April 2026 an independent researcher broke a 15-bit elliptic-curve key on a publicly accessible quantum computer (Project Eleven’s “Q-Day Prize”). The previous public record, in September 2025, was 6 bits. Hardware progress is real but incremental: Google showed error correction below threshold on its 105-qubit Willow chip (late 2024), and IBM reported record logical-qubit fidelity in February 2026, albeit for only 55 microseconds.
Two conclusions matter for planning:
- Elliptic curves may fall first. ECDH and ECDSA (used everywhere in TLS and IKEv2) need fewer logical qubits to break than RSA of comparable classical strength.
- The estimates dropped by about two orders of magnitude in seven years. A planning horizon that assumes “decades” is no longer defensible.
Does that mean you should switch to large MODP groups (DH-4096) instead? No. Finite-field Diffie-Hellman is broken by the same Shor algorithm. A larger group raises the attacker’s cost only polynomially, roughly in proportion to the key size, not exponentially as for classical attacks. It buys a little time at best. Meanwhile it costs CPU time and packet space (512 bytes per DH-4096 public value), and it does nothing against harvest-now-decrypt-later once a CRQC exists. The way out is a post-quantum algorithm, ideally combined with a fast elliptic curve in a PQ/T hybrid (section 3.2).
The symmetric side remains calm. A 2026 analysis by cryptographer Filippo Valsorda argues that because Grover’s algorithm parallelises so poorly, attacking AES-128 would need about 2⁴⁷ parallel quantum machines and some 2¹⁰⁴·⁵ operations in total. That is in line with NIST and BSI, which keep AES-128 acceptable.
1.4 Harvest now, decrypt later
For IPsec the decisive argument is not when a CRQC will exist, but what an adversary can do today:
- Confidentiality is at risk now. An attacker can record encrypted traffic (TLS, SSH, IPsec VPNs) today and decrypt it later, once a CRQC is available (“harvest now, decrypt later”, HNDL). Long-lived data such as health records, intellectual property and state secrets is worth storing.
- Authentication is at risk later. A signature can only be forged after a CRQC exists; a handshake that is already finished cannot be impersonated retroactively. Long-lived trust anchors, certificates and firmware-signing keys still need a head start.
This leads to the core priority for every protocol: make the key exchange quantum-resistant first, then the authentication.
A useful rule of thumb is Mosca’s inequality. If x is how long your data must stay confidential, y is how long your migration takes, and z is the time until a CRQC exists, then you are already too late when x + y > z. For network infrastructure with hardware refresh cycles of 5–10 years, y alone is large.
1.5 Policy deadlines
| Who | Document | Key dates |
|---|---|---|
| NIST (US) | NIST IR 8547 (final, 27 Oct 2026) | Quantum-vulnerable public-key algorithms deprecated after 2030 and disallowed after 2035 |
| EU Member States | Coordinated Implementation Roadmap (June 2025) | Start transition by end of 2026; high-risk use cases migrated by end of 2030; as many systems as feasible by 2035 |
| UK NCSC | Timelines for migration to PQC (Mar 2025) | Plan by 2028; highest-priority migrations by 2031; complete by 2035 |
| Germany (BSI) | TR-02102-1, version 2026-01 | Classical-only key agreement no longer recommended after 2031 (2030 for the most sensitive uses) |
| France (ANSSI) | Announcement, June 2026 | No security certification for products without PQC from 2027 |
| Australia (ASD) | ISM / Planning for PQC (Sep 2025) | Cease traditional asymmetric cryptography by end of 2030 |
| NSA (US NSS) | CNSA 2.0, made binding by CNSSP 15 (Committee on National Security Systems Policy 15) | New acquisitions CNSA 2.0-compliant from 1 Jan 2027; see section 2.4 |
Coloured segments show what each authority expects in that period. Grey hatching marks the past. Hover or tab to a segment for the exact requirement.
Show data as table
| Authority | Period | Requirement |
|---|---|---|
| EU (NIS CG roadmap) | 2025–2027 | EU: national roadmaps and first steps by end of 2026 |
| EU (NIS CG roadmap) | 2027–2031 | EU: high-risk use cases migrated by end of 2030; no stand-alone quantum-vulnerable public-key crypto there after 2030 |
| EU (NIS CG roadmap) | 2031–2036 | EU: as many systems as feasible migrated by 2035 |
| NIST (IR 8547) | 2025–2031 | NIST: transition period; quantum-vulnerable algorithms still allowed until end of 2030 |
| NIST (IR 8547) | 2031–2036 | NIST: quantum-vulnerable public-key algorithms deprecated after 2030 |
| NIST (IR 8547) | 2036–… | NIST: quantum-vulnerable public-key algorithms disallowed after 2035 |
| NSA (CNSSP 15) | 2025–2027 | NSS: prepare; vendors add CNSA 2.0 support |
| NSA (CNSSP 15) | 2027–2031 | NSS: new acquisitions CNSA 2.0-compliant from 1 Jan 2027; non-supporting equipment phased out by 31 Dec 2030 |
| NSA (CNSSP 15) | 2031–2032 | NSS: CNSA 2.0 algorithms mandated by 31 Dec 2031 |
| NSA (CNSSP 15) | 2032–… | NSS: CNSA 2.0 mandatory; all NSS quantum-resistant by 2035 (NSM-10) |
| UK (NCSC) | 2025–2028 | UK: discovery and migration plan by 2028 |
| UK (NCSC) | 2028–2031 | UK: highest-priority migrations by 2031 |
| UK (NCSC) | 2031–2035 | UK: complete migration of all systems by 2035 |
| UK (NCSC) | 2035–… | UK: migration to PQC complete |
| Germany (BSI) | 2025–2031 | BSI TR-02102-1: hybrid PQC key establishment recommended now |
| Germany (BSI) | 2031–2032 | BSI: classical-only key agreement no longer recommended after 2030 for the most sensitive uses |
| Germany (BSI) | 2032–… | BSI: classical-only key agreement no longer recommended after 2031 |
| France (ANSSI) | 2025–2027 | ANSSI: hybrid PQC strongly recommended |
| France (ANSSI) | 2027–… | ANSSI (announced June 2026): no security certification for products without PQC from 2027 |
| Australia (ASD) | 2025–2027 | ASD: refined transition plan by end of 2026 |
| Australia (ASD) | 2027–2029 | ASD: migration of critical systems started by end of 2028 |
| Australia (ASD) | 2029–2031 | ASD: transition complete by end of 2030 |
| Australia (ASD) | 2031–… | ASD: traditional asymmetric cryptography not approved beyond 2030 |
How urgent is it? For any system that protects data with a confidentiality requirement beyond roughly 2030, PQ/T hybrid key exchange should be deployed now. For IPsec, the standards and code points are in place; see the companion article Post-Quantum Cryptography in IPsec.
2. Who sets the rules?
Several kinds of organisations shape PQC. The IETF writes the protocol specifications. NIST standardises the algorithms. National and regional authorities decide which algorithms you may use, in which mode, and by when; for US national security systems, that is the NSA. (For IPsec, IANA also matters: it assigns the numbers that go on the wire, see the IPsec article.)
2.1 The IETF and its RFCs
IETF, Internet-Drafts and RFCs
The Internet Engineering Task Force (IETF) develops Internet protocols in open working groups (WGs), grouped into areas. Security protocols such as IPsec live in the Security Area. The IESG (Internet Engineering Steering Group) approves documents for publication.
Work starts as an Internet-Draft (I-D). A draft named draft-ietf-<wg>-… has been adopted by a working group. A draft named draft-<author>-… is an individual submission with no formal standing. Drafts expire after six months unless they are updated, and they should be cited only as “work in progress.” Typical states you will see on the Datatracker are WG Document, Submitted to IESG for Publication, RFC Ed Queue, and finally RFC.
A Request for Comments (RFC) is never changed after publication. Corrections go into errata, and new versions get a new number that obsoletes or updates the older one. RFCs come from different streams: IETF, IRTF, IAB, and the Independent Submission stream. Independent RFCs are not IETF standards. RFC 9206, the CNSA 1.0 profile for IPsec, is an example.
RFC statuses
| Status | Meaning |
|---|---|
| Proposed Standard | First level of the Standards Track |
| Internet Standard | Second (and final) level; gets an STD number. IKEv2 itself is STD 79 (RFC 7296). |
| Best Current Practice (BCP) | Process or operational guidance |
| Informational | Not a standard: terminology, overviews, guidance (e.g. RFC 9794, RFC 9958) |
| Experimental | Specifications for experimentation |
| Historic | Superseded or obsolete (e.g. IKEv1 since RFC 9395) |
“Proposed Standard” is, in practice, the standard
RFC 2026 originally defined three maturity levels: Proposed Standard, Draft Standard and Internet Standard. RFC 6410 (2011) cut this to two by merging Draft Standard into Internet Standard. RFC 7127 (2014) then rewrote the characterisation of Proposed Standard: such a specification is stable, has been thoroughly reviewed, and is deployed and relied upon on the Internet. Very few documents are ever advanced to Internet Standard.
So when you read that a specification is “only” a Proposed Standard: that is the IETF standard. RFC 9370 (multiple key exchanges), RFC 9242 (IKE_INTERMEDIATE), RFC 8784 (PPK) and the upcoming ML-KEM and ML-DSA specifications for IKEv2 are all on the Standards Track at Proposed Standard level.
2.2 NIST and the FIPS standards
The US National Institute of Standards and Technology (NIST) ran the public PQC competition that started in 2016. The resulting Federal Information Processing Standards (FIPS) are binding for US federal systems. They are also the de-facto global reference, adopted by the IETF and by most national agencies, with some local preferences.
| Publication | Content | Status (Nov 2026) |
|---|---|---|
| FIPS 203 | ML-KEM (from CRYSTALS-Kyber), key encapsulation | Final, 13 Aug 2024 |
| FIPS 204 | ML-DSA (from CRYSTALS-Dilithium), signatures | Final, 13 Aug 2024 |
| FIPS 205 | SLH-DSA (from SPHINCS+), stateless hash-based signatures | Final, 13 Aug 2024 |
| FIPS 206 | FN-DSA (from Falcon), compact lattice signatures | In development; not yet published as a final standard |
| HQC standard | HQC, code-based backup KEM (expected as FIPS 207) | Selected 11 Mar 2025 (NIST IR 8545); standard not yet final |
| SP 800-208 | LMS and XMSS, stateful hash-based signatures | Final (2020) |
| SP 800-227 | Recommendations for KEMs, including combining them | Final, 18 Sep 2025 |
| NIST IR 8547 | Transition to PQC standards (timeline) | Final, 27 Oct 2026 |
| NIST IR 8610 | Status report, round 2 of the additional signature schemes | Final, 14 May 2026 |
What this means in practice:
- ML-KEM is the PQ key exchange used by the IETF for TLS, SSH and IKEv2.
- ML-DSA and SLH-DSA are the PQ signature algorithms for authentication and certificates.
- HQC matters for PQ/PQ hybrids. It uses different mathematics (codes rather than lattices) as a hedge against a break of ML-KEM.
- FIPS validation of cryptographic modules (CMVP, FIPS 140-3) determines when PQC is usable in regulated environments.
2.3 European and Australian authorities
NIST defines the algorithms, but national and regional authorities decide which ones you may use, in which mode (hybrid or not), and by when. Their positions differ, especially on hybrids, so check the guidance for the country and sector your IPsec deployment serves.
European Union
| Body | Role | Key PQC output |
|---|---|---|
| European Commission | Policy | Recommendation on a Coordinated Implementation Roadmap for the transition to PQC (11 April 2024) |
| NIS Cooperation Group, PQC work stream (co-chaired by Germany, France and the Netherlands) | Coordination between Member States | Coordinated Implementation Roadmap for the Transition to PQC (23 June 2025): start by end of 2026, high-risk use cases by end of 2030, as many systems as feasible by 2035. A FAQ followed in September 2026; a second deliverable is in preparation. |
| ENISA | EU cybersecurity agency | Studies on PQC and migration; participates in the roadmap and G7 work |
| ETSI TC CYBER, Quantum-Safe Cryptography WG | European standards body | ETSI TS 103 744 Quantum-safe Hybrid Key Establishment (V1.2.2, Feb 2026): ECDH + ML-KEM combiners (CatKDF, CasKDF). A European Norm on authenticated hybrid key establishment is in development. |
The roadmap’s high-risk category explicitly includes critical infrastructure such as energy, water, health, finance and transport. For those, the guidance is that quantum-vulnerable public-key mechanisms should no longer be used on their own after 2030. In practice that means hybrid or pure PQ.
National authorities in Europe
| Country | Authority | Position and key documents |
|---|---|---|
| Belgium | CCB (Centre for Cybersecurity Belgium) | Co-authored the Cyber Security Coalition white paper series Preparing for the Quantum Era (2025) |
| BIPT (telecom regulator) | Communication of 13 January 2026 on minimum requirements and roadmap for the post-quantum transition in the telecommunications sector: inventory, risk analysis and a phased roadmap for Belgian operators | |
| Germany | BSI | Technical Guideline TR-02102-1 (version 2026-01) recommends hybrid key establishment with ML-KEM-768/1024 or FrodoKEM-976/1344. Classical-only key agreement is no longer recommended after end of 2031 (end of 2030 for the most sensitive applications). In 2026 BSI advised against Classic McEliece for new developments. IPsec specifics are in TR-02102-3. |
| France | ANSSI | Insists on hybridation wherever quantum protection is needed (position papers 2022, 2023; FAQ Oct 2025). Published the technical guide Transition post-quantique d’IPsec (2 Feb 2026), alongside guides for TLS 1.3 and SSHv2, and ANSSI views on crypto agility (Jan 2026). Announced in June 2026 that from 2027 it will no longer certify products without PQC. |
| Netherlands | AIVD, with CWI and TNO | The PQC Migration Handbook, 2nd edition (Dec 2024), plus the PQChoiceAssistant tool |
| United Kingdom | NCSC | Timelines for migration to PQC: discovery and plan by 2028, highest-priority migrations by 2031, full migration by 2035. Treats hybrids as an interim measure; ML-KEM-768 as the general-purpose choice. |
At G7 level, the Cybersecurity Working Group (Canadian presidency 2025, French presidency 2026, with the European Commission and ENISA as guests) issued a joint statement on PQC migration in 2025 and a call to action in September 2026.
Australia
The Australian Signals Directorate (ASD), through its Australian Cyber Security Centre, publishes the Information Security Manual (ISM). The ISM’s Guidelines for Cryptography define the ASD-approved algorithms.
- Deadline: ASD recommends ceasing the use of traditional asymmetric cryptography (RSA, DH, ECDH, ECDSA) by the end of 2030, five years earlier than NIST and the UK. Traditional algorithms are not approved beyond 2030.
- Milestones (Planning for post-quantum cryptography, last updated September 2025): a refined transition plan by end of 2026, migration of critical systems started by end of 2028, and the transition complete by end of 2030.
- Algorithms: the December 2024 ISM update approved ML-KEM and ML-DSA at the highest parameter sets: ML-KEM-1024 and ML-DSA-87.
- Hybrids: ASD does not recommend, but does not prohibit, PQ/T hybrid schemes. This is a notable contrast with BSI and ANSSI.
- Method: the LATICE framework, five high-level phases that run from locating and inventorying traditional asymmetric cryptography to implementing PQC and engaging vendors and stakeholders.
What this means for an IPsec design, where several regimes must be met at once, is covered in the IPsec article.
2.4 NSA: Suite B, CNSA 1.0 and CNSA 2.0
NSA publishes the commercial algorithm suites that US National Security Systems (NSS) must use; CNSSP 15 makes them binding. They matter far beyond the US government, because vendors build their products to them.
Algorithm suites
| Function | Suite B (2005) | CNSA 1.0 (2015/16) | CNSA 2.0 (2022) |
|---|---|---|---|
| Encryption | AES-128, AES-256 | AES-256 | AES-256 |
| Key establishment | ECDH P-256, P-384 | ECDH P-384; DH ≥ 3072; RSA ≥ 3072 | ML-KEM-1024 |
| Signatures | ECDSA P-256, P-384 | ECDSA P-384; RSA ≥ 3072 | ML-DSA-87 |
| Software / firmware signing | – | (as above) | LMS, XMSS (SP 800-208; LMS SHA-256/192 recommended) |
| Hashing | SHA-256, SHA-384 | SHA-384 | SHA-384 or SHA-512 (SHA3-384/512 only for internal hardware functions) |
| IPsec profile | RFC 6379 / RFC 6380 (2011), Historic since RFC 8423 (2018) | RFC 9206 (2022, Independent Submission) | draft-guthrie-cnsa2-ipsec-profile-05 (Oct 2026, individual) |
RFC 9206 defines three CNSA 1.0 IPsec suites: CNSA-GCM-256-ECDH-384, CNSA-GCM-256-DH-3072 and CNSA-GCM-256-DH-4096. They all use IKEv2, AES-256-GCM, and PRF_HMAC_SHA_512 (SHA-512 because of availability, although SHA-384 is the CNSA hash).
CNSA 2.0 and IKEv2
(Details for IPsec, including the CNSA IPsec profiles, are in the IPsec article.) In its CNSA 2.0 FAQ, NSA notes that ML-KEM-1024 cannot simply replace the CNSA 1.0 key exchange in IKEv2 because of message size. It expects to use the IETF approach: a smaller initial key exchange followed by a larger, encrypted one. Concretely, CNSA 1.0 key establishment in IKE_SA_INIT is fortified by ML-KEM-1024 in an additional exchange (RFC 9370). NSA does not require hybrids. It rejects non-standard hybrids on NSS, but it does mention PPKs (RFC 8784) as possibly appropriate for specialised cases.
Timeline
Bars: from "support & prefer CNSA 2.0" to "use exclusively" (2022 category table). Dashed lines: CNSSP 15 milestones used by NSA since FAQ v2.1. Grey hatching marks the past. Hover for details.
Show data as table
| Category | Support & prefer by | Exclusively by |
|---|---|---|
| Software / firmware signing | 2025 | 2030 |
| Browsers, servers, cloud | 2025 | 2033 |
| Networking (VPNs, routers) | 2026 | 2030 |
| Operating systems | 2027 | 2033 |
| Niche equipment | 2030 | 2033 |
| Custom & legacy | – | update or replace by 2033 |
In 2022 NSA published the category table shown as bars above. Traditional networking equipment (VPNs, routers) has one of the tightest schedules: support and prefer CNSA 2.0 by 2026, use it exclusively by 2030. Since the December 2024 version of the FAQ (v2.1), NSA has framed the transition around the CNSSP 15 milestones (the dashed lines) rather than the category table. The latest FAQ is v2.2 (October 2026). For IPsec vendors and VPN operators in that ecosystem, 1 January 2027 is the procurement gate. VPN equipment without ML-KEM-1024 support will be increasingly hard to sell or deploy there.
Why CNSA 2.0 accepts only ML-KEM-1024
The CNSA 2.0 FAQ lists ML-KEM-1024 “for all classification levels” (and likewise ML-DSA-87, AES-256 and SHA-384/512). NSA does not publish a detailed rationale beyond that, but the choice follows the same logic as its earlier suites and its other choices:
- One parameter set for everything up to TOP SECRET. CNSA 1.0 already dropped Suite B’s two levels (P-256/AES-128 for SECRET, P-384/AES-256 for TOP SECRET) in favour of the strongest option only. A single set simplifies product certification, configuration and interoperability across all national security systems.
- Maximum security margin for long-lived secrets. Classified information can need protection for decades. Category 5 leaves the most room if cryptanalysis of lattices improves.
- No hybrid required, so the PQ part must stand alone. Because NSA does not require a traditional component, ML-KEM alone has to carry the security. That argues for the largest parameter set.
- The cost is acceptable. ML-KEM-1024 is still fast; its main cost is size (1,568 bytes), which IKEv2 handles with RFC 9242 and RFC 7383.
Australia’s ASD made the same choice (ML-KEM-1024, ML-DSA-87). For most other uses, NIST, the IETF and the UK NCSC consider ML-KEM-768 (category 3) sufficient.
CNSA 2.0 vs. NIST SP 800-227 on PQ/T hybrids
Is there a conflict between NSA’s position and NIST’s KEM guidance? In short: no direct contradiction, but different emphasis.
| Aspect | NIST SP 800-227 (Sep 2025) | NSA CNSA 2.0 (FAQ v2.1/v2.2) |
|---|---|---|
| Hybrids | Accommodates them. It gives guidance on combining the shared secrets of several KEMs or key-establishment schemes into one key with an approved key-derivation method. | Does not require them. Accepts standards-based hybrids (and plans one for IKEv2), rejects non-standard hybrids on NSS. |
| Which PQ component counts | Any approved KEM parameter set (ML-KEM-512, -768, -1024) | Only ML-KEM-1024 |
| Role of the traditional component | Part of the combined key; the result is approved as long as an approved component is combined in an approved way | Adds no CNSA 2.0 credit; in IKEv2 it is kept mainly for practical reasons (message size, interoperability) |
| Long term | NIST IR 8547: quantum-vulnerable algorithms disallowed after 2035 | CNSA 2.0 algorithms mandated by end of 2031 |
For an IPsec designer the practical outcome is the same: ECDH P-384 + ML-KEM-1024 via RFC 9370 satisfies SP 800-227, CNSA 2.0, BSI and ANSSI at once. One open point remains: how long a traditional component may stay inside a hybrid after the 2031 and 2035 dates. Check the final CNSA 2.0 IPsec profile (draft-guthrie-cnsa2-ipsec-profile) when it is published.
3. Concepts: pure PQ, hybrids, hybrid signatures and crypto agility
3.1 Terminology
The IETF vocabulary is defined in RFC 9794 (Terminology for Post-Quantum Traditional Hybrid Schemes, June 2025):
- Traditional algorithm: an asymmetric algorithm based on factorisation or (elliptic-curve) discrete logarithms (RSA, DH, ECDH, ECDSA, EdDSA).
- Post-quantum algorithm: an asymmetric algorithm designed to be secure against both classical and quantum computers (ML-KEM, ML-DSA, SLH-DSA, …). RFC 9794 prefers “post-quantum” over “quantum-resistant” or “quantum-safe”, because the label describes a design goal, not a guarantee.
- Multi-algorithm scheme: a scheme with more than one component algorithm serving the same purpose.
3.2 Pure PQ, PQ/T hybrid and PQ/PQ hybrid
| Approach | Definition | IKEv2 example (details) |
|---|---|---|
| Pure PQ | Only post-quantum components | ML-KEM alone in IKE_SA_INIT. draft-ietf-ipsecme-ikev2-mlkem allows ML-KEM-512 there; ML-KEM-768/1024 SHOULD NOT be used in IKE_SA_INIT over UDP when the path MTU is unknown, because IKE_SA_INIT cannot be fragmented. |
| PQ/T hybrid | At least one traditional and at least one PQ component; designed so that an attacker must break all components | ECDH (P-384 or Curve25519) in IKE_SA_INIT plus ML-KEM in IKE_INTERMEDIATE via RFC 9370 |
| PQ/PQ hybrid | Only PQ components, preferably based on different hard problems | ML-KEM plus FrodoKEM (draft-ietf-ipsecme-hybrid-kem-ikev2-frodo) or plus HQC (draft-reddy-ipsecme-ikev2-hqc, individual) |
RFC 9794 also separates composite from non-composite constructions:
- Composite: the protocol stays the same and the cryptographic element changes. One “algorithm” internally contains several components, for example one composite public key and one composite signature.
- Non-composite: each component keeps its own format, and the protocol changes to carry several of them. IKEv2 with RFC 9370 is RFC 9794’s own example of a non-composite construction.
Why PQ/T hybrid rather than PQ-only?
ML-KEM and ML-DSA are young compared with ECDH and ECDSA. The underlying lattice problems have been studied for far less time, and implementations of new algorithms have bugs (timing side channels such as KyberSlash were found and fixed in early Kyber/ML-KEM implementations in 2024). A hybrid stays secure as long as one component holds:
| Scenario | PQ-only | PQ/T hybrid | Traditional only |
|---|---|---|---|
| No CRQC, PQ algorithm sound | Secure | Secure | Secure |
| No CRQC, PQ algorithm or implementation broken | Broken | Secure (traditional part holds) | Secure |
| CRQC exists, PQ algorithm sound | Secure | Secure (PQ part holds) | Broken |
| CRQC exists and PQ algorithm broken | Broken | Broken | Broken |
A hybrid therefore protects against the two separate risks: a quantum computer, and a flaw in the new algorithm. Once a CRQC exists, the traditional component no longer adds anything, and only the PQ component protects the session. That is why hybrids are seen as a transition measure, and why the PQ component must be strong in its own right. Combined with crypto agility (section 3.4), a hybrid lets you replace a broken PQ algorithm without losing protection in the meantime.
Several European agencies (e.g. BSI and ANSSI) recommend PQ/T hybrids for the transition (section 2.3). Australia’s ASD does not recommend them, but does not prohibit them either. NSA does not require hybrids for CNSA 2.0, but anticipates a hybrid-style construction in IKEv2 for practical reasons (section 2.4).
A note on pre-shared keys: RFC 8784 and RFC 9867 mix a Post-quantum Preshared Key (PPK) into the IKEv2 key derivation (PPK notifications). This is a symmetric hedge against quantum attacks, not a PQ algorithm. It works today with any key exchange, provided you can distribute high-entropy PPKs out of band. RFC 8784 protects the Child SAs but not the initial IKE SA. RFC 9867 (November 2025) moves the PPK into IKE_INTERMEDIATE so that the initial IKE SA, including identities, is protected too. It also allows PPKs that change frequently, for example keys delivered by QKD.
3.3 Hybrid signatures: parallel, serial and composite
For authentication there are three main ways to combine a traditional and a PQ signature. RFC 9794 names them, and RFC 9955 (Hybrid Signature Spectrums, July 2026) analyses their security properties.
| Type | How it works | Properties | IKEv2 work |
|---|---|---|---|
| Parallel signatures | Two independent signatures over the same data, each with its own key and usually its own certificate (“multi-certificate authentication”, a “PQ/T parallel PKI” in RFC 9794 terms) | Simple and backward compatible; the components are separable, so an attacker can strip one and present the other as a valid single signature | Individual drafts, e.g. draft-hu-ipsecme-pqt-hybrid-auth-with-sep-cert, draft-wang-ipsecme-multi-auth-ikev2-pq |
| Serial (nested) signatures | The second signature also covers the first: sig₂ = Sign₂(m ‖ sig₁) | Stripping the outer signature is detectable, stripping the inner one is not; order and verification logic matter | No dedicated IKEv2 specification |
| Composite signatures | One composite key, one composite signature and one algorithm identifier; internally ML-DSA plus a traditional algorithm, bound together by a domain separator | Looks like a single algorithm to the protocol; designed to resist stripping | RFC 10073 (composite ML-DSA, Nov 2026) + draft-ietf-ipsecme-pq-hybrid-auth |
RFC 9955’s central concept is separability: can an attacker turn a hybrid signature into a valid signature for only one component? Its non-separability spectrum runs from fully separable (parallel) to weak and strong non-separability.
3.4 Crypto agility
Crypto agility is the ability to change algorithms without redesigning the protocol or replacing the equipment (see RFC 7696, BCP 201). IKEv2 is agile by design: every algorithm is a negotiated transform with an IANA number, which is why ML-KEM fits in with only three new code points. In practice, agility also requires:
- an inventory of where which algorithms and keys are used (VPN gateways, clients, PKI, HSMs);
- algorithms that are configurable rather than hard-coded, and firmware that can be upgraded;
- room for larger messages: IKE fragmentation (RFC 7383), IKE_INTERMEDIATE (RFC 9242), or TCP encapsulation (RFC 9329);
- downgrade protection. As long as traditional-only proposals are accepted, an active attacker with a CRQC could force them. Once all peers support ML-KEM, policy should refuse proposals without it. A general fix is in progress (
draft-ietf-ipsecme-ikev2-downgrade-prevention).
4. Security categories, parameter sets and sizes
4.1 The five categories
NIST defines five post-quantum security strength categories. Each one is tied to the cost of attacking a well-understood symmetric primitive, under classical and quantum attack:
| Category | At least as hard to break as … |
|---|---|
| 1 | Exhaustive key search on AES-128 |
| 2 | Collision search on SHA-256 |
| 3 | Exhaustive key search on AES-192 |
| 4 | Collision search on SHA-384 |
| 5 | Exhaustive key search on AES-256 |
4.2 Parameter sets per category
| Algorithm (standard) | Cat. 1 | Cat. 2 | Cat. 3 | Cat. 4 | Cat. 5 |
|---|---|---|---|---|---|
| ML-KEM (FIPS 203) | ML-KEM-512 | – | ML-KEM-768 | – | ML-KEM-1024 |
| ML-DSA (FIPS 204) | – | ML-DSA-44 | ML-DSA-65 | – | ML-DSA-87 |
| SLH-DSA (FIPS 205) | 128s, 128f | – | 192s, 192f | – | 256s, 256f |
| FN-DSA (FIPS 206, not final) | FN-DSA-512 | – | – | – | FN-DSA-1024 |
| HQC (expected FIPS 207, not final) | HQC-1 | – | HQC-3 | – | HQC-5 |
| FrodoKEM (not NIST; IETF draft) | FrodoKEM-640 | – | FrodoKEM-976 | – | FrodoKEM-1344 |
No current algorithm targets category 4. SLH-DSA comes in SHA2 and SHAKE variants for each parameter set; the s variants have smaller signatures, the f variants sign faster. In practice, ML-KEM-768 (category 3) is the common default, while CNSA 2.0 and the Australian ISM require category 5 (ML-KEM-1024, ML-DSA-87).
4.3 Key and signature sizes
| Algorithm | Public key (bytes) | Ciphertext / signature (bytes) |
|---|---|---|
| X25519 (traditional, for comparison) | 32 | 32 |
| ECDSA P-256 (traditional, for comparison) | 64 | ≈ 64–72 |
| ECDSA P-384 (traditional, for comparison) | 96 | ≈ 96–104 |
| ECDSA P-521 (traditional, for comparison) | 132 | ≈ 132–139 |
| RSA-3072 (traditional, for comparison) | 384 | 384 |
| RSA-4096 (traditional, for comparison) | 512 | 512 |
| ML-KEM-512 / -768 / -1024 | 800 / 1,184 / 1,568 | 768 / 1,088 / 1,568 |
| ML-DSA-44 / -65 / -87 | 1,312 / 1,952 / 2,592 | 2,420 / 3,309 / 4,627 |
| SLH-DSA-128s / -128f | 32 | 7,856 / 17,088 |
| SLH-DSA-192s / -192f | 48 | 16,224 / 35,664 |
| SLH-DSA-256s / -256f | 64 | 29,792 / 49,856 |
| FN-DSA-512 / -1024 (not final) | 897 / 1,793 | ≈ 666 / ≈ 1,280 |
| FrodoKEM-640 / -976 / -1344 | 9,616 / 15,632 / 21,520 | 9,720 / 15,744 / 21,632 |
PQ keys and signatures are one to three orders of magnitude larger than traditional ones. For protocols that run over UDP, like IKEv2, that is the main engineering challenge; see key exchange and signature sizes in IKEv2.
5. IETF work beyond IPsec: lamps and pquip
The IPsec-specific work happens in the ipsecme working group; see the IPsec article. Two other working groups produce results that every protocol, including IPsec, depends on.
5.1 lamps: Limited Additional Mechanisms for PKIX and SMIME
IKEv2 authentication relies on X.509 certificates, and LAMPS defines how PQ keys and signatures go into them.
| Document | Subject | Status |
|---|---|---|
| RFC 9881 | ML-DSA in X.509 certificates and CRLs | Proposed Standard (Oct 2025) |
| RFC 9909 | SLH-DSA in X.509 | Proposed Standard (Dec 2025) |
| RFC 10073 | Composite ML-DSA (18 combinations of ML-DSA-44/65/87 with RSA, ECDSA, Ed25519/Ed448) | Proposed Standard (Nov 2026), formerly draft-ietf-lamps-pq-composite-sigs |
5.2 pquip: Post-Quantum Use In Protocols
PQUIP does not define protocols. It produces cross-cutting guidance, published as Informational RFCs.
| Document | Subject | Status |
|---|---|---|
| RFC 9794 | Terminology for PQ/T hybrid schemes | Informational (Jun 2025) |
| RFC 9958 | Post-Quantum Cryptography for Engineers, formerly draft-ietf-pquip-pqc-engineers | Informational (Jun 2026) |
| RFC 9955 | Hybrid signature spectrums | Informational (Jul 2026) |
draft-ietf-pquip-hbs-state | Hash-based signatures: state and backup management (LMS/XMSS) | Active with the IESG |
draft-ietf-pquip-pqc-hsm-constrained | Adapting constrained devices (HSMs, embedded) for PQC | Active with the IESG |
draft-prabel-pquip-pqc-overview-00 | Overview of PQ algorithms (parameters, sizes, assumptions) | Individual draft, expired Sep 2026; no formal standing |
If you only read one document, read RFC 9958. It explains the threat model, KEMs versus Diffie-Hellman, signature trade-offs, hybrids, key and signature sizes, and migration in engineering terms.
6. Sources
IETF: RFCs and drafts
- RFC 2026, RFC 6410, RFC 7127: IETF standards process
- RFC 7696: Guidelines for Cryptographic Algorithm Agility (BCP 201)
- RFC 9794: Terminology for PQ/T Hybrid Schemes
- RFC 9958: Post-Quantum Cryptography for Engineers
- RFC 9955: Hybrid Signature Spectrums
- RFC 9881: ML-DSA in X.509
- RFC 9909: SLH-DSA in X.509
- RFC 10073: Composite ML-DSA
- RFC 6379, RFC 6380 (Suite B for IPsec, Historic), RFC 8423, RFC 9206 (CNSA for IPsec)
- pquip, active with the IESG: draft-ietf-pquip-hbs-state, draft-ietf-pquip-pqc-hsm-constrained
- Individual: draft-guthrie-cnsa2-ipsec-profile, draft-prabel-pquip-pqc-overview (expired)
- Working group pages: pquip · lamps
NIST
- NIST Post-Quantum Cryptography project and publications
- NIST SP 800-227: Recommendations for Key-Encapsulation Mechanisms
- NIST IR 8547 (final)
NSA
European and Australian authorities
- European Commission: Post-Quantum Cryptography policy page
- NIS Cooperation Group: Coordinated Implementation Roadmap (June 2025) and survey feedback / FAQ (Sep 2026)
- ETSI TS 103 744 V1.2.2: Quantum-safe Hybrid Key Establishment
- CCB: Preparing for the Quantum Era
- BIPT: Communication of 13 January 2026 on the post-quantum transition in telecommunications
- ANSSI: Post-quantum cryptography and Transition post-quantique d’IPsec (Feb 2026)
- AIVD, CWI, TNO: The PQC Migration Handbook, 2nd edition
- UK NCSC: Timelines for migration to post-quantum cryptography
- ASD: Planning for post-quantum cryptography
Research and news on quantum computing
- C. Gidney, How to factor 2048 bit RSA integers with less than a million noisy qubits (2025)
- Live Science / Yahoo: Quantum computers might need just 10,000 qubits to break encryption (Mar 2026), on the Caltech/Oratomic preprint arXiv:2603.28627
- Interesting Engineering: Caltech and Oratomic error-correction architecture (Mar 2026)
- Discover Magazine: three recent quantum computing breakthroughs (Apr 2026)
- Live Science: IBM logical-qubit fidelity record (Apr 2026)
- The Block: 15-bit elliptic-curve key broken, Project Eleven Q-Day Prize (Apr 2026)
- The Quantum Insider: AES-128 likely safe from quantum attack (Apr 2026)
Comments
No comments yet.