Flexible Hybrid PQ MLS Combiner

Flexible Hybrid PQ MLS Combiner AWS

alwenjo@amazon.com

Naval Postgraduate School

britta.hale@nps.edu

AWS

mulmarta@amazon.ch

Naval Postgraduate School

xisen.tian1@nps.edu

Security Messaging Layer Security security authenticated key exchange PCS Post-Quantum This document describes a protocol for combining a traditional MLS session with a post-quantum (PQ) MLS session to achieve flexible and efficient hybrid PQ confidentiality and authenticity that amortizes the computational cost of PQ Key Encapsulation Mechanisms and Digital Signature Algorithms. Specifically, we describe how to use the exporter secret of a PQ MLS session, i.e., an MLS session using a PQ ciphersuite, to seed PQ guarantees into an MLS session using a traditional ciphersuite. By supporting on-demand traditional-only key updates (a.k.a. PARTIAL updates) or hybrid-PQ key updates (a.k.a. FULL updates), we can reduce the bandwidth and computational overhead associated with PQ operations while meeting the requirement of frequent key rotations. About This Document The latest revision of this draft can be found at . Status information for this document may be found at . Discussion of this document takes place on the Messaging Layer Security Working Group mailing list (), which is archived at . Subscribe at . Source for this draft and an issue tracker can be found at .

Introduction A fully capable quantum adversary has the ability to break fundamental underlying cryptographic assumptions of traditional Key Encapsulation Mechanisms (KEMs) and Digital Signature Algorithms (DSAs). This has led to the development of post-quantum (PQ) cryptographically secure KEMs and DSAs by the cryptographic research community which have been formally adopted by the National Institute of Standards and Technology (NIST), including the Module Lattice KEM (ML-KEM) and Module Lattice DSA (ML-DSA) algorithms. While these provide PQ security, ML-KEM and ML-DSA have significant overhead in terms of public key size, signature size, ciphertext size, and CPU time compared to their traditional counterparts. This has made achieving PQ entity and data authenticity particularly challenging. The hybrid approach in this draft amortizes the PQ overhead costs enabling practical PQ confidentiality or PQ confidentiality and PQ authenticity. Moreover, research arms on side-channel attacks, etc., have motivated uses of hybrid-PQ combiners that draw security from both the underlying PQ and underlying traditional components. A variety of hybrid security treatments have arisen across IETF working groups to bridge the gap between performance and security to encourage the adoption of PQ security in existing protocols, including the MLS protocol . Within the MLS working group, there are various ways to approach PQ security extensions:

A single MLS ciphersuite for a hybrid post-quantum/traditional KEM. The ciphersuite can act as a drop-in replacement for the KEM, focusing on hybrid confidentiality but not authenticity, and does not incur changes elsewhere in the MLS stack. As a confidentiality focus, it addresses the the harvest-now / decrypt-later threat model. However, every key epoch incurs a PQ overhead cost.
Mechanisms that leverage hybridization as a means to not only address the security balance between PQ and traditional components and achieve resistance to harvest-now / decrypt-later attacks, but also use it as a means to improve performance of PQ use while achieving PQ authenticity as well.

This document addresses the second topic of these work items.

Terminology The key words "MUST", "MUST NOT", "REQUIRED", "SHALL", "SHALL NOT", "SHOULD", "SHOULD NOT", "RECOMMENDED", "NOT RECOMMENDED", "MAY", and "OPTIONAL" in this document are to be interpreted as described in BCP 14 when, and only when, they appear in all capitals, as shown here. The terms MLS client, MLS member, MLS group, Leaf Node, GroupContext, KeyPackage, Signature Key, Handshake Message, Private Message, Public Message, and RequiredCapabilities have the same meanings as in the MLS protocol .

Notation We use terms from from MLS and PQ Hybrid Terminology . Below, we have restated relevant terms and define new ones: Application Message: : A PrivateMessage carrying application data. Handshake Message: : A PublicMessage or PrivateMessage carrying an MLS Proposal or Commit object, as opposed to application data. Key Derivation Function (KDF): : A Hashed Message Authentication Code (HMAC)-based expand-and-extract key derivation function (HKDF) as described in . Key Encapsulation Mechanism (KEM): : A key transport protocol that allows two parties to obtain a shared secret based on the receiver's public key. Post-Quantum (PQ) MLS Session: : An MLS session that uses a PQ-KEM construction, such as described by FIPS 203 from NIST. It may optionally also use a PQ-DSA construction, such as described by FIPS 204 from NIST. Traditional MLS Session: : An MLS session that uses a Diffie-Hellman (DH) based KEM as described in . PQ/T: : A Post-Quantum and Traditional hybrid (protocol).

The Combiner Protocol Execution The hybrid PQ MLS (HPQMLS) combiner protocol runs two MLS sessions in parallel, synchronizing their group memberships. The two sessions are combined by exporting a secret from the PQ session and importing it as a Pre-Shared Key (PSK) into the traditional session. This combination process is mandatory for Commits of Add and Remove proposals in order to maintain synchronization between the sessions. However, it is optional for any other Commits (e.g. to allow for less computationally expensive traditional key rotations). Due to the higher computational costs and output sizes of PQ KEM (and signature) operations, it may be desirable to issue PQ combined (a.k.a. FULL) Commits less frequently than the traditional-only (a.k.a. PARTIAL) Commits. Since FULL Commits introduce PQ security into the MLS key schedule, the overall key schedule remains PQ-secure even when PARTIAL Commits are used. The FULL Commit rate establishes the post-quantum Post-Compromise Security (PCS) window, while the PARTIAL Commit rate can tighten the traditional PCS window even while maintaining PQ security more generally. The combiner protocol design treats both sessions as black-box interfaces so we only highlight operations requiring synchronizations in this document. The default way to start a HPQMLS combined session is to create a PQ MLS session and then start a traditional MLS session with the exported PSK from the PQ session, as previously mentioned. Alternatively, a combined session can also be created after a traditional MLS session has already been running. This is done through creating a PQ MLS session with the same group members, sending a Welcome message containing the HPQMLSInfo struct in the GroupContext, and then making a FULL Commit as described in in the section.

Commit Flow Commits to proposals MAY be PARTIAL or FULL. For a PARTIAL Commit, only the traditional session's epoch is updated following the Propose-Commit sequence from Section 12 of . For a FULL Commit, a Commit is first applied to the PQ session and another Commit is applied to the traditional session using a PSK derived from the PQ session using the DeriveExtensionSecret and hpqmls_psk label (see ). To ensure the correct PSK is imported into the traditional session, the sender includes information about the PSK in a PreSharedKey proposal for the traditional session's Commit list of proposals. The information about the exported PSK is captured (shown '=' in the figures below for illustration purposes) by the PreSharedKeyID struct as detailed in . Receivers process the PQ Commit to derive a new epoch in the PQ session and then the traditional Commit (which also includes the PSK proposal) to derive the new epoch in the traditional session. | | | | | | Commit'() | | | Commit(PreSharedKeyID) | |<--------------------------------------------------------------------+ | |<---------------------------+ Fig 1a. FULL Commit to an empty proposal list. Messages with ' are sent in the the PQ session. PreSharedKeyID identifies a PSK exported from the PQ session in the new epoch following a Commit'(). The PreSharedKeyID is implicitly included in the commit in the classical session via the PreSharedKey Proposal. ]]> | | | | | | Upd'(B) | | | Upd(B, f) | |<------------------------------------------------------------------------+ | |<-------------------------------+ | | | | Commit'(Upd') | | | PresharedKeyID = | | | DeriveExtensionSecret('hpqmls_psk') | | | Commit(Upd, PreSharedKeyID) | | |------------------------------------------------------------------------>| | | | | | Commit'(Upd') | | | Commit(Upd, PreSharedKeyID) | |<------------------------------------------------------------------------+ | |<-------------------------------+ Fig 1b. FULL Commit to an Update proposal from Client B. Messages with ' are sent in the the PQ session. ]]>

Adding a User User leaf nodes are first added to the PQ session following the sequence described in Section 3 of except using PQ algorithms where HPKE algorithms exist. For example, a PQ-DSA signed PQ KeyPackage, i.e. containing a PQ public key, must first be published via the Distribution Service (DS). Then the associated Commit and Welcome messages will be sent and processed in the PQ session according to Section 12 of . The same sequence is repeated in the standard session except following the FULL Commit combining sequence where a PreSharedKeyID proposal is additionally committed. The joiner MUST issue a FULL Commit as soon as possible after joining to achieve PCS. | | | | | | Welcome' | | | | Welcome(PreSharedKeyID) | | | +----------------------------------------->| | | | | | | | | | Commit'(Add'(KeyPackageB')) | | | | Commit(Add(KeyPackageB), PreSharedKeyID) | |<----------------------------------------------------------------------------------------------------+ Figure 2: Client A adds client B to the group. Messages with ' come from the PQ session. Processing Welcome and Commit in the traditional sessio requires the PSK exported exported from the PQ session. ]]>

Welcome Message Validation Since a client must join two sessions, the Welcome messages it receives to each session must indicate that it is not sufficient to join only one or the other. Therefore, the HPQMLSInfo struct indicating the GroupID and ciphersuites of the two sessions MUST be included in the Welcome message via serialization as a GroupContext Extension in order to validate joining the combined sessions. All members MUST verify group membership is consistent in both sessions after a join and the new member MUST issue a FULL Commit as described in Fig 1b.

External Joins External joins are used by members who join a group without being explicitly added (via an Add-Commit sequence) by another existing member. The external user MUST join both the PQ session and the traditional session. As stated previously, the GroupInfo used to create the External Commit MUST contain the HPQMLSInfo struct. After joining, the new member MUST issue a FULL Commit as described in Fig 1b.

Removing a Group Member User removals MUST be done in both PQ and traditional sessions followed by a FULL Commit Update as as described in Fig 1b. Members MUST verify group membership is consistent in both sessions after a removal.

Application Messages HPQMLS combiner provides PQ security to the traditional MLS session. Application messages are therefore only sent in the traditional session using the encryption_secret provided by the key schedule of the traditional session according to Section 15 of .

Modes of Operation Security needs vary by organizations and system-specific risk tolerance and/or constraints. While this combiner protocol targets combining a PQ session and a traditional session the degree of PQ security may be tuned depending on the use-case: i.e., as PQ/T Confidentiality Only or both PQ/T Confidentiality and PQ/T Authenticity. For PQ/T Confidentiality Only, the PQ session MUST use a PQ KEM, while for PQ authenticity, the PQ session MUST use both a PQ KEM and a PQ DSA. The modes of operation are specified by the mode flag in HPQMLSInfo struct and are listed below.

PQ/T Confidentiality Only The default mode of operation is PQ/T Confidentiality Only mode. This mode provides confidentiality and limited authenticity against quantum attackers. More precisely, it provides PQ authenticity against "outsiders", that is, against quantum attackers who do not have acces to (signature) secret keys of any group member. (Authenticity comes from the fact that the traditional session adds AEAD / MAC tags which are not available to outsiders with CRQC.) This mode does not prevent quantum impersonation attacks by other group members. That is, a group member with a CRQC can successfully impersonate another group member. Note that an active attacker with access to a CRQC can become a group member by impersonating members in the moment they are added. As such, the authenticity guarantees outlined above only hold as long as the adversary is passive during the addition of new group members.

PQ/T Confidentiality + Authenticity The elevated mode of operation is the PQ/T Confidentiality + Authenticity mode. Under a use environment of a cryptographically relevant quantum computer (CRQC), the threat model used in the default mode would be too weak and assurance about update authenticity is required. Recall that authenticity in MLS refers to three types of guarantees: 1) that messages were sent by a member of the group provided by the computed symmetric group key used in AEAD, 2) that key updates were performed by a valid member of the group, and 3) that a message was sent by a particular user (i.e. non-repudiation) provided by digital signatures on messages. While the symmetric group key used for AEAD in the traditional session remains protected from a CRQC adversary through the PSK from the PQ session, signatures would not be secure against forgery without using a PQ DSA to sign handshake messages nor are application messages assured to have non-repudiation against a CRQC adversary. Therefore, in the PQ/T Confidentiality + Authenticity mode, the PQ session MUST use a PQ DSA in addition to PQ KEM ciphersuites for handshake messages (the traditional session remains unchanged). This version of PQ authenticity provides PQ authenticity to the PQ session's MLS commit messages, strengthening assurance for (1) and ensuring (2). These in turn provide PQ assurance for the key schedule from which application keys are derived in the traditional session. Application keys are used in an AEAD for protection of MLS application messages and thereby inherit the PQ security. However, it should be noted that PQ non-repudation security for application messages as described by (3) is not achieved by this mode. Achieving PQ non-repudiation on application messages would require hybrid signatures in the traditional session, with considerations to options described in .

Extension Requirements to MLS The HPQMLSInfo struct contains characterizing information to signal to users that they are participating in a hybrid session. This is necessary both functionally to allow for group synchronization and as a security measure to prevent downgrading attacks to coax users into parcipating in just one of the two sessions. The group_id, cipher_suite, and epoch from both sessions (t for the traditional session and pq for the PQ session) are used as bookkeeping values to validate and synchronize group operations. The mode is a boolean value: 0 for the default PQ/T Confidentiality Only mode and 1 for the PQ/T Confidentiality + Authenticity mode. The HPQMLSInfo struct conforms to the Safe Extensions API (see ). Recall that an extension is called safe if it does not modify base MLS protocol or other MLS extensions beyond using components of the Safe Extension API. This allows security analysis of our HPQMLS Combiner protocol in isolation of the security guarantees of the base MLS protocol to enable composability of guarantees. The HPMLSInfo extension struct SHALL be in the following format: ; } ExtensionContent; struct{ opaque t_session_group_id; opaque PQ_session_group_id; bool mode; CipherSuite t_cipher_suite; CipherSuite pq_cipher_suite; uint64 t_epoch; uint64 pq_epoch; } HPQMLSInfo ]]>

Key Schedule The hpqmls_psk exporter key derived in the PQ session MUST be derived in accordance with the Safe Extensions API guidance (see Exporting Secrets in ). In particular, it SHALL NOT use the extension_secret and MUST be derived from only the epoch_secret from the key schedule in . This is to ensure forward secrecy guarantees (see ). Even though the hpqmls_psk PSK is not sent over the wire, members of the HPQMLS session must agree on the value of which PSK to use. In alignment with the Safe Extensions API policy for PSKs, HPQMLS PSKs used SHALL set PSKType = 3 and extension_type = HPQMLS (see Section 2.1.6 Pre-Shared Keys in ). --> KDF.Extract [...] | | +--> DeriveSecret(., "welcome") | = welcome_secret | V ExpandWithLabel(., "epoch", GroupContext_[n], KDF.Nh) | | V epoch_secret | | +--> DeriveSecret(., ) | = [...] Fig 3: The hpqmls_psk of the PQ session is injected into the key schedule of the traditional session using the safe extensions API DeriveExtensionSecret. ]]>

Cryptographic Objects

Cipher Suites There are no changes to how cipher suites are used to perform group key computations from RFC9420. However, the choice of which primitives are used by the traditional and PQ subsessions must be explicitly stated by the CipherSuite objects within HPQMLSInfo. So long as the traditional session only uses classical primitives and the PQ session uses PQ primitives for KEM, a HPQMLS session is valid. Specifically, the PQ primitives for HPQMLS must be 'pure' (fully) PQ: PQ cost is already being amoritized at the protocol level so allowing hybrid PQ cipher suites to be used in the PQ session only adds extra overhead and complexity. Furthermore, the pq_cipher_suite may contain a classical digital signature algorithm used if mode is set to 0 (PQ Confidentiality-Only) but MUST be fully PQ if mode is set to 1 (PQ Confidentiality+Authenticity). These cipher suite combinations and modes MUST not be toggled or modified after a HPQMLS session has commenced. Clients MUST reject a HPQMLS session with invalid or duplicate cipher suites (e.g. two traditional cipher suites).

Key Encapsulation Mechanism For HPQMLS sessions, the PQ subsession MUST use a Key Encapsulation Mechanism (KEM) that is standardized by NIST for post-quantum cryptography. Specifically, only KEMs that have been selected and published by NIST as part of their post-quantum cryptography standardization process (e.g., ML-KEM as specified in FIPS 203) are permitted for use in the PQ session. The use of experimental, non-standardized, or hybrid KEMs in the PQ session is NOT RECOMMENDED and MUST be rejected by compliant clients. This requirement ensures interoperability and a consistent security baseline across all HPQMLS deployments.

Signing For HPQMLS sessions, the choice of digital signature algorithm in the PQ subsession depends on the selected mode of operation. If the mode is set to 1 (PQ Confidentiality+Authenticity), the PQ session MUST use a digital signature algorithm that is standardized by NIST for post-quantum cryptography, such as ML-DSA as specified in FIPS 204. The use of experimental, non-standardized, or hybrid signature algorithms in the PQ session is NOT RECOMMENDED and MUST be rejected by compliant clients in this mode. If the mode is set to 0 (PQ Confidentiality-Only), the PQ session MAY use a classical digital signature algorithm, but the use of a NIST-standardized PQ signature algorithm is RECOMMENDED. These requirements ensure that the authenticity guarantees of HPQMLS sessions are aligned with the intended security level and provide a consistent baseline for interoperability and security across deployments.

Security Considerations

FULL Commit Frequency So long as the FULL Commit flow is followed for group administration actions, PQ security is extended to the traditional session. Therefore, FULL Commits can occur as frequently or infrequently as desired by any given security policy. This results in a flexible and efficient use of compute, storage, and bandwidth resources for the host by mainly calling partial updates on the traditional MLS session, given that the group membership is stable. Thus, our protocol provides PQ security and can maintain a tighter PCS window against traditional attackers as well as forward secrecy window against traditional or quantum attackers with lower overhead when compared to running a single MLS session that only uses PQ KEMs or PQ KEM/DSAs. Furthermore, the PQ PCS window against quantum attackers can be selected based on an application and even variable over time, ranging from e.g. a single FULL Commit in PQ/T Confidentiality Only mode followed by PARTIAL Commits from that point onwards (enabling general PQ/traditional confidentiality, traditional update authenticity, traditional PCS, and PQ/traditional forward secrecy) to frequent FULL Commits in the same mode (enabling general PQ/traditional confidentiality, traditional update authenticity, PQ/traditional PCS, and PQ/traditional forward secrecy). In PQ/T Confidentiality + Authenticity mode with frequent FULL Commits, the latter case would enable general PQ/traditional confidentiality, PQ/traditional update authenticity, PQ/traditional PCS, and PQ/traditional forward secrecy.

Attacks on Non-Repudiation While PQ message integrity is provided by the symmetric key used in AEAD, attacks on non-repudiation (e.g., source forgery) on application messages may still be possible by a CRQC adversary since only traditional signatures on used after the AEAD. However, in terms of group key agreement, this is insufficient to mount anything more than a denial-of-service attack (e.g. via group state desynchronization). In terms of application messages, a traditional DSA signature may be forged by an external CRQC adversary, but the content (including sender information) is still protected by AEAD which uses the symmetric group key. Thus, an external CRQC adversary can only conduct a false-framing attack, where group members are assured of the authenticity of a message being sent by a group member for the adversary has changed the signature to imply a different sender; it would require an insider CRQC adversary to actually mount a masquerading or forgery attack, which is beyond the scope of this protocol. If this is a concern, hybrid PQ DSAs can be used in the traditional session to sign application messages. Since this would negate much of the efficiency gains from using this protocol and denial-of-service attacks can be achieve through more expeditious means, such a option is not considered here.

Forward Secrecy Recall that one of the ways MLS achieves forward secrecy is by deleting security sensitive values after they are consumed (e.g. to encrypt or derive other keys/nonces) and the key schedule has entered a new epoch. For example, values such as the init_secret or epoch_secret are deleted at the start of a new epoch. If the MLS exporter_secret or the extension_secret from the PQ session is used directly as a PSK for the traditional session, against the requirements set above, then there is a potential scenario in which an adversary can break forward secrecy because these keys are derived during an epoch and are not deleted. Therefore, the hpqmls_psk MUST be derived from the epoch_secret created at the start of an epoch from the PQ session (see Figure 3) to ensure forward secrecy.

Transport Security Recommendations for preventing denial-of-service attacks or restricting transmitted messages are inherited from MLS.

IANA Considerations The MLS sessions combined by this protocol conform to the IANA registries listed for MLS .

Normative References The Messaging Layer Security (MLS) Protocol Messaging applications are increasingly making use of end-to-end security mechanisms to ensure that messages are only accessible to the communicating endpoints, and not to any servers involved in delivering messages. Establishing keys to provide such protections is challenging for group chat settings, in which more than two clients need to agree on a key but may not be online at the same time. In this document, we specify a key establishment protocol that provides efficient asynchronous group key establishment with forward secrecy (FS) and post-compromise security (PCS) for groups in size ranging from two to thousands. Key words for use in RFCs to Indicate Requirement Levels In many standards track documents several words are used to signify the requirements in the specification. These words are often capitalized. This document defines these words as they should be interpreted in IETF documents. This document specifies an Internet Best Current Practices for the Internet Community, and requests discussion and suggestions for improvements. Ambiguity of Uppercase vs Lowercase in RFC 2119 Key Words RFC 2119 specifies common key words that may be used in protocol specifications. This document aims to reduce the ambiguity by clarifying that only UPPERCASE usage of the key words have the defined special meanings. Terminology for Post-Quantum Traditional Hybrid Schemes UK National Cyber Security Centre UK National Cyber Security Centre Naval Postgraduate School One aspect of the transition to post-quantum algorithms in cryptographic protocols is the development of hybrid schemes that incorporate both post-quantum and traditional asymmetric algorithms. This document defines terminology for such schemes. It is intended to be used as a reference and, hopefully, to ensure consistency and clarity across different protocols, standards, and organisations. HMAC-based Extract-and-Expand Key Derivation Function (HKDF) This document specifies a simple Hashed Message Authentication Code (HMAC)-based key derivation function (HKDF), which can be used as a building block in various protocols and applications. The key derivation function (KDF) is intended to support a wide range of applications and requirements, and is conservative in its use of cryptographic hash functions. This document is not an Internet Standards Track specification; it is published for informational purposes. Hybrid Public Key Encryption This document describes a scheme for hybrid public key encryption (HPKE). This scheme provides a variant of public key encryption of arbitrary-sized plaintexts for a recipient public key. It also includes three authenticated variants, including one that authenticates possession of a pre-shared key and two optional ones that authenticate possession of a key encapsulation mechanism (KEM) private key. HPKE works for any combination of an asymmetric KEM, key derivation function (KDF), and authenticated encryption with additional data (AEAD) encryption function. Some authenticated variants may not be supported by all KEMs. We provide instantiations of the scheme using widely used and efficient primitives, such as Elliptic Curve Diffie-Hellman (ECDH) key agreement, HMAC-based key derivation function (HKDF), and SHA2. This document is a product of the Crypto Forum Research Group (CFRG) in the IRTF. Hybrid signature spectrums SandboxAQ Naval Postgraduate School SandboxAQ UK National Cyber Security Centre This document describes classification of design goals and security considerations for hybrid digital signature schemes, including proof composability, non-separability of the component signatures given a hybrid signature, backwards/forwards compatiblity, hybrid generality, and simultaneous verification. Discussion of this work is encouraged to happen on the IETF PQUIP mailing list pqc@ietf.org or on the GitHub repository which contains the draft: https://github.com/dconnolly/draft-hale-pquip-hybrid- signature-spectrums The Messaging Layer Security (MLS) Extensions Phoenix R&D The Messaging Layer Security (MLS) protocol is an asynchronous group authenticated key exchange protocol. MLS provides a number of capabilities to applications, as well as several extension points internal to the protocol. This document provides a consolidated application API, guidance for how the protocol's extension points should be used, and a few concrete examples of both core protocol extensions and uses of the application API.

Acknowledgments

Contributors Konrad Kohbrok Phoenix R&D Email: konrad.kohbrok@datashrine.de