Site-to-site and remote access VPNs carry years of sensitive traffic that an adversary can record today and decrypt once a cryptographically relevant quantum computer exists. IPsec tunnels negotiated with IKEv2 and classical Diffie-Hellman are the textbook "harvest now, decrypt later" target, and NIST's draft transition plan (IR 8547) names IKE among the interactive protocols that should move to post-quantum key establishment first. The IETF has already built the plumbing: RFC 9370 lets IKEv2 run several key exchanges in one negotiation, RFC 8784 mixes post-quantum preshared keys into the key schedule, and an ML-KEM specification for IKEv2 is in the RFC Editor queue. This guide explains how those pieces fit, how to test hybrid proposals without dropping tunnels, and what to ask VPN vendors. It is for network security engineers, VPN owners and architects. For the organisation-wide programme, see the Post-Quantum Cryptography Migration Guide.
Key takeaways
- Confidentiality first, authentication later. RFC 9370 and RFC 8784 protect tunnel keys against recorded-traffic decryption, but neither makes RSA or ECDSA peer authentication quantum-resistant.
- Hybrid is the practical default. Run a classical group such as X25519 or ECP-384 in IKE_SA_INIT, then ML-KEM as an additional key exchange in IKE_INTERMEDIATE, so the keys stay safe if either algorithm holds.
- Packet size is the main operational risk. ML-KEM-768 and ML-KEM-1024 messages can exceed the path MTU, so put them after IKE_SA_INIT and enable IKEv2 fragmentation (RFC 7383).
- PPKs work today on many stacks. RFC 8784 post-quantum preshared keys give quantum-resistant tunnel keys without new algorithms, at the price of out-of-band key distribution and rotation.
- Check draft status before you depend on it. The ML-KEM for IKEv2 document was draft-ietf-ipsecme-ikev2-mlkem-09 in the RFC Editor queue when checked on 27 October 2026; code points 35, 36 and 37 are assigned, but read the final RFC when it appears.
Why VPN traffic is an early migration target
IPsec protects some of the longest-lived traffic an organisation sends: links between data centres, branch offices, cloud VPCs and partners, often carrying database replication, backups and management traffic. Two properties make it urgent:
- Recordability: encrypted tunnel traffic crosses networks the organisation does not control and can be captured passively. If the IKE key exchange is broken later, every Child SA key derived from it falls, and so does the recorded payload.
- Long equipment lifetimes: VPN concentrators, firewalls and routers stay in service for many years. A box bought today without a post-quantum path may still be terminating tunnels after 2035, when the NIST IR 8547 draft proposes to disallow quantum-vulnerable key establishment.
Authentication is a lower immediate risk for VPNs. A forged signature only helps an attacker who has a quantum computer at the time of the handshake, whereas broken key exchange helps anyone who recorded traffic earlier. That is why the standards so far tackle key establishment first. The general background on IPsec modes and IKE is in the IPsec Security Guide.
The building blocks: RFC 9242, RFC 9370 and ML-KEM
| Document | Status (checked 27 October 2026) | What it adds |
|---|---|---|
| RFC 7296 | Internet Standard | IKEv2 base protocol: IKE_SA_INIT, IKE_AUTH, CREATE_CHILD_SA |
| RFC 7383 | Standards Track | IKEv2 message fragmentation, needed for large encrypted payloads |
| RFC 9242 | Standards Track | IKE_INTERMEDIATE exchange between IKE_SA_INIT and IKE_AUTH; its messages are encrypted and can be fragmented |
| RFC 9370 (May 2023) | Standards Track, updates RFC 7296 | Up to seven additional key exchanges (transform types ADDKE1 to ADDKE7), IKE_FOLLOWUP_KE exchange for rekeys |
| RFC 8784 (June 2020) | Standards Track | Post-quantum preshared keys (PPKs) mixed into SK_d, SK_pi and SK_pr |
| draft-ietf-ipsecme-ikev2-mlkem-09 (5 July 2026) | Internet-Draft, in RFC Editor queue | ML-KEM-512, ML-KEM-768 and ML-KEM-1024 as key exchange methods 35, 36 and 37, alone or as additional exchanges |
How a hybrid IKEv2 negotiation runs
- IKE_SA_INIT: peers negotiate the usual Transform Type 4 key exchange (for example X25519 or ECP-384) plus one or more ADDKE transforms, and exchange the INTERMEDIATE_EXCHANGE_SUPPORTED notification.
- IKE_INTERMEDIATE: each additional key exchange runs in its own encrypted IKE_INTERMEDIATE exchange, in ascending ADDKE order. After each one, a new SKEYSEED is derived from the previous SK_d and the new shared secret, so the final keys depend on every exchange.
- IKE_AUTH: peers authenticate as before (certificates, PSK or EAP), now under keys that include the post-quantum secret.
- Rekeying: CREATE_CHILD_SA is followed by IKE_FOLLOWUP_KE exchanges carrying the additional key exchanges, linked by the ADDITIONAL_KEY_EXCHANGE notification. New SAs exist only after the whole series completes.
RFC 9370 is backward compatible: if a peer does not support IKE_INTERMEDIATE, the ADDKE transforms are treated as unknown and skipped, and an initiator can offer NONE for an ADDKE type to make the post-quantum round optional. That property is what makes a staged rollout possible, and also what makes silent downgrade possible if you never enforce it. RFC 9370 also notes the childless IKE SA mode (RFC 6023), which avoids creating the first Child SA under keys not yet protected by every exchange.
Choosing ML-KEM parameter sets
| Option | NIST category | Where it fits in IKEv2 | Trade-off |
|---|---|---|---|
| X25519 + ML-KEM-768 | 3 | ML-KEM in IKE_INTERMEDIATE | Common hybrid pairing; needs fragmentation support on the path |
| ECP-384 + ML-KEM-1024 | 5 | ML-KEM in IKE_INTERMEDIATE | Higher assurance profile for national security style requirements; largest messages |
| ML-KEM-512 alone or first | 1 | The draft says it is unlikely to exceed typical path MTUs and may be used in IKE_SA_INIT | Lowest security margin; useful where fragmentation is impossible |
| Classical + RFC 8784 PPK | Depends on PPK entropy | No new algorithm; PPK mixed into the key schedule | Key distribution and rotation burden; initial IKE SA keys not protected |
The draft advises that implementations sending IKE over UDP without path MTU discovery, or where the PMTU is unknown, should not put ML-KEM-768 or ML-KEM-1024 in IKE_SA_INIT, because unencrypted IKE_SA_INIT cannot be fragmented. Later exchanges can carry them using RFC 7383 fragmentation.
Post-quantum preshared keys (RFC 8784)
RFC 8784 is the option many teams can use on existing equipment. Both peers hold a long, high-entropy secret, the PPK, identified by a PPK_ID. The initiator signals USE_PPK (notify 16435) in IKE_SA_INIT, sends PPK_IDENTITY (16436) in IKE_AUTH, and the PPK is mixed into SK_d, SK_pi and SK_pr, so all Child SA keys depend on it.
- Optional mode: the initiator also sends NO_PPK_AUTH (16437) with authentication computed without the PPK, so peers that lack it can fall back. Use this during rollout only.
- Mandatory mode: once every peer is upgraded, set PPKs mandatory. The RFC explains that with optional PPKs and signature authentication, a quantum-capable active attacker could strip USE_PPK and forge signatures to downgrade the session.
- Limits: the initial IKE SA encryption keys are not protected, so a future quantum attacker could see peer identities and negotiated parameters, though not Child SA keys. The RFC recommends against group PPKs, because one stolen PPK exposes every tunnel using it.
PPKs are secrets like any other. Store them in a secrets manager or HSM-backed store, give each peer pair its own PPK, and rotate them on a schedule. See the Secrets Management Guide.
Where identity fits
A VPN gateway is a non-human identity: it authenticates with a certificate or a preshared key, and in remote access it validates user and device credentials. Post-quantum key exchange does not change that authentication. Three identity tasks still belong in the plan:
- Gateway certificates: RSA and ECDSA certificates used in IKE_AUTH fall under the 2035 disallowance in the IR 8547 draft. Inventory them, avoid issuing gateway certificates with validity past 2035 on quantum-vulnerable algorithms, and track vendor support for ML-DSA authentication. See the Machine Identity, PKI and Certificate Lifecycle Guide.
- Preshared keys and PPKs: treat them as machine credentials with owners, rotation dates and storage requirements, not as configuration text in a firewall backup.
- Management plane: administrator and API access to VPN appliances is often more exposed than the tunnels. Post-quantum tunnels do not help if management credentials leak.
Rolling out hybrid IKEv2
- Inventory tunnels and peers. List every IKEv2 tunnel with peer type, software version, current proposals, authentication method, data sensitivity and data confidentiality lifetime. Flag remaining IKEv1 tunnels: they need upgrading before any of this applies.
- Ask vendors for evidence. Request support status for RFC 9242, RFC 9370, RFC 8784, ML-KEM code points 35 to 37 and RFC 7383 fragmentation, by firmware version, plus FIPS 140-3 status of the module providing ML-KEM.
- Lab test interoperability. Pair your gateway with each peer type in a lab. Open-source stacks such as strongSwan express a hybrid proposal with additional key exchange keywords, for example x25519-ke1_mlkem768, and use keX_none to make a round optional.
- Test the path. Confirm fragmented IKE messages survive firewalls, NAT devices and cloud load balancers between peers. Capture IKE_INTERMEDIATE exchanges and check for retransmissions and timeouts.
- Pilot in optional mode. Enable the hybrid proposal (or optional PPK) on a small set of low-risk tunnels, with the classical proposal still accepted. Monitor negotiation results for two full rekey cycles.
- Prioritise by data lifetime. Move tunnels carrying long-lived confidential data (replication, backups, partner data, regulated records) first.
- Enforce. Once a peer group negotiates hybrid reliably, remove the classical-only proposal or set PPKs mandatory, so a downgrade fails rather than succeeds quietly.
- Roll back cleanly. Keep the previous proposal set in version control and a documented change to re-add classical-only proposals per tunnel if a peer fails after an upgrade.
Remote access clients and cloud VPN endpoints
Site-to-site tunnels are usually the first candidates because both ends are under one team's control. Remote access is harder for three reasons:
- Client diversity: built-in operating system IKEv2 clients, vendor agents and mobile clients each gain RFC 9370 support on their own schedule. A gateway can only enforce hybrid once every supported client negotiates it, so expect a long optional phase and track client versions as part of the inventory.
- Hostile networks: hotel, mobile and guest networks are where fragmented UDP is most often dropped. Prefer configurations that keep IKE_SA_INIT small and rely on IKE_INTERMEDIATE fragmentation, and test from real consumer networks before wide rollout.
- Managed cloud gateways: cloud provider VPN services expose a fixed list of supported proposals. Check the provider documentation for post-quantum options per region and record the gap where none exists, because the far end of a hybrid tunnel must support the same ADDKE transform.
Where a remote access platform cannot yet offer post-quantum key exchange, reduce the amount of long-lived sensitive data crossing it, for example by moving administrative access to application-level brokers with their own protected transport.
Decision guidance
- Hybrid ML-KEM vs PPK: choose hybrid ML-KEM where both peers support RFC 9370 and the path handles fragmentation; it needs no extra secret distribution. Choose PPK where equipment supports RFC 8784 but not ML-KEM, or as defence in depth for the most sensitive links.
- Category 3 vs 5: ML-KEM-768 is the common balance of size and margin; ML-KEM-1024 suits environments that must meet higher assurance profiles and can tolerate larger exchanges.
- Small organisations: most will depend on managed firewall or SASE providers. The task is to get a dated support statement for hybrid IKEv2 and include it in renewal decisions.
- Large estates: standardise a small number of proposal profiles, manage them centrally, and track compliance per tunnel rather than per device.
- Replace vs wait: if a platform has no roadmap for RFC 9370 or ML-KEM and its support life runs past 2030, plan replacement in the next refresh cycle rather than at end of life.
Detection and measurement
Log IKE negotiation results on every gateway: the selected Transform Type 4 group, any ADDKE transforms selected, PPK use, fragmentation, retransmissions and NO_PROPOSAL_CHOSEN failures. Feed these into the same pipeline as other VPN telemetry, described in the VPN and Firewall Log Detection Guide. Alert when a tunnel that previously negotiated a hybrid exchange falls back to classical only, which can indicate a configuration regression or an active downgrade.
| Metric | Target direction | Why it matters |
|---|---|---|
| Percentage of tunnels negotiating a post-quantum exchange (ML-KEM or PPK) | Rising to 100% for high-sensitivity tunnels | Core progress measure |
| Tunnels with classical-only proposals still accepted | Falling | Shows downgrade exposure |
| IKE negotiation failure and retransmission rate after change | Stable versus baseline | Detects fragmentation and interop problems |
| Gateways on firmware without RFC 9370 support | Falling | Drives refresh planning |
| PPKs past rotation date | Zero | Shows PPKs are managed as secrets |
| Remaining IKEv1 tunnels | Zero | Prerequisite for every post-quantum option |
Framework mapping
| Framework | Reference | Relevance |
|---|---|---|
| NIST SP 800-53 Rev. 5 | SC-8 and SC-8(1) Transmission Confidentiality and Integrity, Cryptographic Protection | Tunnel encryption strength |
| NIST SP 800-53 Rev. 5 | SC-12 Cryptographic Key Establishment and Management; SC-13 Cryptographic Protection | Key exchange algorithms and PPK handling |
| NIST SP 800-53 Rev. 5 | AC-17 Remote Access | Remote access VPN policy |
| NIST CSF 2.0 | PR.DS-02, PR.IR-01, PR.PS-01 | Data in transit, network protection, configuration management of proposals |
| NIST SP 800-77 Rev. 1 | Guide to IPsec VPNs | Baseline IPsec and IKE deployment guidance |
| NIST IR 8547 (ipd) | Section 4.2 application-specific standards | IKE named for priority post-quantum key establishment |
Common failure patterns
- Optional forever: hybrid proposals offered alongside classical ones and never enforced give no protection against an active downgrade.
- ML-KEM in IKE_SA_INIT over constrained paths: large unencrypted messages get dropped or fragmented at IP level, causing intermittent tunnel failures that look like network faults.
- One PPK for every peer: a group PPK turns one leaked firewall backup into exposure of all tunnels, which RFC 8784 warns against.
- Assuming authentication is covered: teams report "quantum-safe VPN" while gateways still authenticate with RSA certificates valid beyond 2035.
- Testing only the lab path: cloud VPN endpoints and carrier NAT behave differently from lab switches with fragmented UDP.
- Ignoring the appliance itself: post-quantum tunnels on an unpatched VPN appliance leave the larger risk untouched. See the VPN Security Guide.
Practitioner checklist
- Inventory every IKE tunnel with peer, firmware, proposals, authentication method and data lifetime.
- Retire remaining IKEv1 tunnels.
- Obtain dated vendor statements on RFC 9242, RFC 9370, RFC 8784, ML-KEM (IDs 35 to 37) and RFC 7383 support.
- Lab test hybrid proposals with every peer type, including fragmented exchanges across real network paths.
- Place ML-KEM-768 or ML-KEM-1024 in IKE_INTERMEDIATE, not IKE_SA_INIT, unless the path MTU is guaranteed.
- Use per-peer PPKs stored in a secrets manager, rotated on a schedule, and set to mandatory after rollout.
- Enforce hybrid once stable and alert on any fallback to classical-only negotiation.
- Track gateway certificate algorithms and validity against the 2035 line.
- Re-check the ML-KEM for IKEv2 document status and publish the RFC number in your standards once issued.
Standards and references
- IETF: RFC 9370 Multiple Key Exchanges in IKEv2
- IETF: RFC 8784 Mixing Preshared Keys in IKEv2 for Post-quantum Security
- IETF: Post-quantum Key Exchange with ML-KEM in IKEv2 (Internet-Draft)
- NIST: SP 800-77 Rev. 1 Guide to IPsec VPNs
- NIST: IR 8547 (Initial Public Draft) Transition to Post-Quantum Cryptography Standards
- NIST: FIPS 203 Module-Lattice-Based Key-Encapsulation Mechanism Standard
- strongSwan: Proposals documentation
Related NHI Mgmt Group resources: Post-Quantum Cryptography Migration Guide · NIST IR 8547 PQC Transition Timeline Guide · Hybrid Post-Quantum TLS Key Exchange Guide · IPsec Security Guide