Internet-Draft | ietf-pquip-hybrid-spectrums | November 2024 |
Bindel, et al. | Expires 9 May 2025 | [Page] |
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 compatibility, 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-ietf-pquip-hybrid-signature-spectrums¶
This note is to be removed before publishing as an RFC.¶
Discussion of this document takes place on the Post-Quantum Use In Protocols Working Group mailing list (pqc@ietf.org), which is archived at https://mailarchive.ietf.org/arch/browse/pqc/.¶
Source for this draft and an issue tracker can be found at https://github.com/dconnolly/draft-connolly-pquip-hybrid-signature-spectrums.¶
This Internet-Draft is submitted in full conformance with the provisions of BCP 78 and BCP 79.¶
Internet-Drafts are working documents of the Internet Engineering Task Force (IETF). Note that other groups may also distribute working documents as Internet-Drafts. The list of current Internet-Drafts is at https://datatracker.ietf.org/drafts/current/.¶
Internet-Drafts are draft documents valid for a maximum of six months and may be updated, replaced, or obsoleted by other documents at any time. It is inappropriate to use Internet-Drafts as reference material or to cite them other than as "work in progress."¶
This Internet-Draft will expire on 9 May 2025.¶
Copyright (c) 2024 IETF Trust and the persons identified as the document authors. All rights reserved.¶
This document is subject to BCP 78 and the IETF Trust's Legal Provisions Relating to IETF Documents (https://trustee.ietf.org/license-info) in effect on the date of publication of this document. Please review these documents carefully, as they describe your rights and restrictions with respect to this document. Code Components extracted from this document must include Revised BSD License text as described in Section 4.e of the Trust Legal Provisions and are provided without warranty as described in the Revised BSD License.¶
The initial focus on the transition to use of post-quantum algorithms in protocols has largely been on confidentiality, given the potential risk of store and decrypt attacks, where data encrypted today using traditional algorithms could be decrypted in the future by an attacker with a Cryptographically-Relevant Quantum Computer (CRQC). While traditional authentication is only at risk once a CRQC exists, it is important to consider the transition to post-quantum authentication before this point. This is particularly relevant for systems where algorithm turn-over is complex or takes a long time (e.g., long-lived systems with hardware roots of trust), or where future checks on past authenticity play a role (e.g., digital signatures on legal documents).¶
The relative newness of many (although not all) post-quantum algorithms means that less cryptanalysis of such algorithms is available than for long-established counterparts, such as RSA and elliptic-curve based solutions for confidentiality and authenticity. This has drawn attention to hybrid cryptographic schemes, which combine both traditional and post-quantum (or more generally next-generation) algorithms in one cryptographic scheme. These may offer increased assurance for implementers, namely that as long as the security of one of the two component algorithms of the hybrid scheme holds, the confidentiality or authenticity offered by that scheme is maintained.¶
Whether or not hybridization is desired depends on the use case and security threat model. Conservative users may not have complete trust in the post-quantum algorithms or implementations available, while also recognizing a need to start post-quantum transition. For such users, hybridization can support near-term transition while also avoiding trusting solo post-quantum algorithms too early. On the other hand, hybrid schemes, particularly for authentication, may introduce significant complexity into a system or a transition process, so might not be the right choice for all. For cases where hybridization is determined to be advantageous, a decision on how to hybridize needs to be made. With many options available, this document is intended to provide context on some of the trade-offs and nuances to consider.¶
Hybridization has been looked at for key encapsulation [HYBRIDKEM], and in an initial sense for digital signatures [HYBRIDSIG]. Compared to key encapsulation, hybridization of digital signatures, where the verification tag may be expected to attest to both standard and post-quantum components, is subtler to design and implement due to the potential separability of the hybrid/dual signatures and the risk of downgrade/stripping attacks. There are also a range of requirements and properties that may be required from hybrid signatures, not all of which can be achieved at once.¶
This document focuses on explaining advantages and disadvantages of different hybrid signature scheme designs and different security goals for them. It is intended as a resource for designers and implementers of hybrid signature schemes to help them decide what properties they do and do not require from their scheme. It does not attempt to answer the question of whether or not a hybrid scheme is desirable for, or should be used in a given case. It also intentionally does not propose concrete hybrid signature combiners or instantiations thereof.¶
We follow existing Internet drafts on hybrid terminology [I-D.ietf-pquip-pqt-hybrid-terminology] and hybrid key encapsulation mechanisms (KEM) [I-D.ietf-tls-hybrid-design] to enable settling on a consistent language. We will make clear when this is not possible. In particular, we follow the definition of 'post-quantum algorithm', 'traditional algorithms', and 'combiner'. Moreover, we use the definition of 'certificate' to mean 'public-key certificate' as defined in [RFC4949].¶
Signature scheme: A signature scheme is defined via the following three algorithms:¶
KeyGen() -> (pk, sk)
: A probabilistic key generation algorithm,
which generates a public verifying key pk
and a secret signing key
sk
.¶
Sign(sk, m) -> (sig)
: A probabilistic signature generation, which
takes as input a secret signing key sk
and a message m
, and
outputs a signature sig
.¶
Verify(pk, sig, m) -> b
: A verification algorithm, which takes as
input a public verifying key pk
, a signature sig
and a message
m
, and outputs a bit b
indicating accept (b=1)
or reject
(b=0)
of the signature for message m
.¶
Hybrid signature scheme: Following [I-D.ietf-pquip-pqt-hybrid-terminology], we define a hybrid signature scheme to be "a multi-algorithm digital signature scheme made up of two or more component digital signature algorithms ...". While it often makes sense for security purposes to require that the security of the component schemes is based on the hardness of different cryptographic assumptions, in other cases hybrid schemes might be motivated, e.g., by interoperability of variants on the same scheme and as such both component schemes are based on the same hardness assumption (e.g., both post-quantum assumptions or even both the same concrete assumption such as Ring LWE). We allow this explicitly. This means in particular that in contrast to [I-D.ietf-pquip-pqt-hybrid-terminology], we will use the more general term 'hybrid signature scheme' instead of requiring one post-quantum and one traditional algorithm (i.e., PQ/T hybrid signature schemes) to allow also the combination of several post-quantum algorithms. The term 'composite scheme' is sometimes used as a synonym for 'hybrid scheme'. This is different from [I-D.ietf-pquip-pqt-hybrid-terminology] where the term is used as a specific instantiation of hybrid schemes such that "where multiple cryptographic algorithms are combined to form a single key or signature such that they can be treated as a single atomic object at the protocol level." To avoid confusing we will avoid the term 'composite scheme'.¶
Hybrid signature: A hybrid signature is the output of the hybrid signature scheme's signature generation. As synonyms we might use 'dual signature'. For example, NIST define a dual signature as "two or more signatures on a common message" [NIST_PQC_FAQ]. For the same reason as above we will avoid using the term 'composite signature' although it sometimes appears as synonym for 'hybrid/dual signature'.¶
Component (signature) scheme: Component signature schemes are the cryptographic algorithms contributing to the hybrid signature scheme. This has a similar purpose as in [I-D.ietf-pquip-pqt-hybrid-terminology]. 'Ingredient (signature) scheme' may be used as a synonym.¶
Next-generation algorithms: Following [I-D.ietf-tls-hybrid-design], we define next-generation algorithms to be "algorithms which are not yet widely deployed but which may eventually be widely deployed". Hybrid signatures are mostly motivated by preparation for post-quantum transition, hence the reference to post-quantum algorithms through this draft. However, the majority of the discussion in this document applies equally well to future transitions to other next-generation algorithms.¶
Artifact: An artifact is evidence of the sender's intent to hybridize a signature that remains even if a component algorithm tag is removed. Artifacts can be e.g., at the algorithmic level (e.g., within the digital signature), or at the protocol level (e.g., within the certificate), or on the system policy level (e.g., within the message). Artifacts should be easily identifiable by the receiver in the case of signature stripping.¶
Before diving into the design goals for hybrid digital signatures, it is worth taking a look at why hybrid digital signatures are desirable for some applications. As many of the arguments hold in general for hybrid algorithms, we again refer to [I-D.ietf-tls-hybrid-design] that summarizes these well. In addition, we explicate the motivation for hybrid signatures here.¶
Next-generation algorithms and their underlying hardness assumptions are often more complex than traditional algorithms. For example, the signature scheme ML-DSA (a.k.a. CRYSTALS-Dilithium) that has been selected for standardization by NIST. While the scheme follows the well-known Fiat-Shamir transform to construct the signature scheme, it also relies on rejection sampling that is known to give cache side channel information (although this does not lead to a known attack). Likewise, the signature scheme Falcon uses complex sampling during signature generation. Furthermore, recent attacks again the next-generation multivariate schemes Rainbow and GeMSS might call into question the asymptotic and concrete security for conservative adopters and therefore might hinder adoption.¶
As such, some next-generation algorithms carry a higher risk of implementation mistakes and revision of parameters compared to traditional algorithms, such as RSA. RSA is a relatively simple algorithm to understand and explain, yet during its existence and use there have been multiple attacks and refinements, such as adding requirements to how padding and keys are chosen, and implementation issues such as cross-protocol attacks. Thus, even in a relatively simple algorithm subtleties and caveats on implementation and use can arise over time. Given the complexity of next generation algorithms, the chance of such discoveries and caveats needs to be taken into account.¶
Of note, some next generation algorithms have received substantial analysis attention, for example through the NIST Post-Quantum Cryptography Standardization Process [NIST_PQC_FAQ]. Thus, if and when further information on caveats and implementation issues come to light, it is less likely that a "break" will be catastrophic. Instead, such vulnerabilities and issues may represent a weakening of security - which may in turn be offset if a hybrid approach has been used. The complexity of post-quantum algorithms needs to be balanced against the fact that hybridization itself adds more complexity to a protocol and introduces the risk of implementation mistakes in the hybridization process.¶
One example of a next generation algorithm is the signature scheme ML-DSA (a.k.a. CRYSTALS-Dilithium) that has been selected for standardization by NIST. While the scheme follows the well-known Fiat-Shamir transform to construct the signature scheme, it also relies on rejection sampling that is known to give cache side channel information (although this does not lead to a known attack). Furthermore, recent attacks again the post-quantum multivariate schemes Rainbow and GeMSS might call into question the asymptotic and concrete security for conservative adopters and therefore might hinder adoption.¶
The need to transition to post-quantum algorithms now while
simultaneously being aware of potential, hidden subtleties in their
resistance to standard attacks drives transition designs towards
hybridization. Mosca’s equation [MOSCA] very simply illustrates the
risk of post-quantum transition delay: l + d > q
, where l is the
information life-span, d is the time for system transition to
post-quantum algorithms, and q is the time before a quantum computer is
ready to execute cryptanalysis. As opposed to key exchange and KEMs, it
may not be obvious why there is urgency for an adoption of post-quantum
signatures; namely, while encryption is subject to
store-now-decrypt-later attacks, there may not seem to be a parallel
notion for authenticity, i.e., 'store-now-modify-later attacks'.¶
However, in larger systems, including national systems, space systems, large healthcare support systems, and critical infrastructure, where acquisition and procurement time can be measured in years and algorithm replacement may be difficult or even practically impossible, this equation can have drastic implications. In such systems, algorithm turn-over can be complex and difficult and can take considerable time (such as in long-lived systems with hardware deployment), meaning that an algorithm may be committed to long-term, with no option for replacement. Long-term commitment creates further urgency for immediate post-quantum algorithm selection. Additionally, for some sectors future checks on past authenticity plays a role (e.g., many legal, financial, auditing, and governmental systems). The 'store-now-modify-later' analogy would present challenges in such sectors, where future analysis of past authentication may be more critical than in e.g., internet connection use cases. As such there is an eagerness to use post-quantum signature algorithms for some applications.¶
There are various security goals that can be achieved through hybridization. The following provides a summary of these goals, while also noting where security goals are in conflict, i.e., that achievement of one goal precludes another, such as backwards compatibility.¶
One goal of hybrid signature schemes is security. As defined in [I-D.ietf-pquip-pqt-hybrid-terminology], ideally a hybrid signature scheme can achieve 'hybrid authentication' which is the property that (cryptographic) authentication is achieved by the hybrid signature scheme provided that a least one component signature algorithm remains 'secure'. There might be, however, other goals in competition with this one, such as backward-compatibility. Hybrid authentication is an umbrella term that encompasses more specific concepts of hybrid signature security, such as 'hybrid unforgeability' described next.¶
Hybrid unforgeability is a specific type of hybrid authentication, where the security assumption for the scheme, e.g. EUF-CMA or SUF-CMA, is maintained as long as at least one of the component schemes is EUF-CMA (resp., SUF-CMA) secure without a prioritisation. We call this notion 'hybrid unforgeability'; it is a specific type of hybrid authentication. For example, the concatenation combiner in [HYBRIDSIG] is 'hybrid unforgeable'. As mentioned above, this might be incompatible with backward-compatibility, where the EUF-CMA (resp., SUF-CMA) security of the hybrid signature relies solely on the security of one of the component schemes instead of relying on both, e.g., the dual message combiner using nesting in [HYBRIDSIG]. For more details, we refer to our discussion below.¶
Use cases where a hybrid scheme is used with, e.g., EUF-CMA security assumed for only one component scheme generally use hybrid techniques for their functional transition pathway support, while fully trusting either the traditional or post-quantum algorithm. In contrast, use cases where a hybrid scheme is used with e.g., EUF-CMA security assumed for both component schemes without prioritisation can use hybrid techniques for both functional transition and security transition, where it may not be known which algorithm should be relied upon.¶
Under proof composability, the component algorithms are combined in such a way that it is possible to prove a security reduction from the security properties of a hybrid signature scheme to the properties of the respective component signature schemes and, potentially, other building blocks such as hash functions, KDF, etc. Otherwise an entirely new proof of security is required, and there is a lack of assurance that the combination builds on the standardization processes and analysis performed to date on component algorithms. The resulting hybrid signature would be, in effect, an entirely new algorithm of its own. The more the component signature schemes are entangled, the more likely it is that an entirely new proof is required, thus not meeting proof composability.¶
Non-Separability was one of the earliest properties of hybrid digital signatures to be discussed [HYBRIDSIG]. It was defined as the guarantee that an adversary cannot simply “remove” one of the component signatures without evidence left behind. For example there are artifacts that a carefully designed verifier may be able to identify, or that are identifiable in later audits. This was later termed Weak Non-Separability (WNS) [HYBRIDSIGDESIGN]. Note that WNS does not restrict an adversary from potentially creating a valid component digital signature from a hybrid one (a signature stripping attack), but rather implies that such a digital signature will contain artifacts of the separation. Thus authentication that is normally assured under correct verification of digital signature(s), is now potentially also reliant on further investigation on the receiver side that may extend well beyond traditional signature verification behavior. For instance, this can intuitively be seen in cases of a message containing a context note on hybrid authentication, that is then signed by all component algorithms/the hybrid signature scheme. If an adversary removes one component signature but not the other, then artifacts in the message itself point to the possible existence of hybrid signature such as a label stating “this message must be hybrid signed”. This might be a counter measure against stripping attacks if the verifier expects a hybrid signature scheme to have this property. However, it places the responsibility of signature validity not only on the correct format of the message, as in a traditional signature security guarantee, but the precise content thereof.¶
Strong Non-Separability (SNS) is a stronger notion of WNS, introduced in
[HYBRIDSIGDESIGN]. SNS guarantees that an adversary cannot take as input a
hybrid signature (and message) and output a valid component signature (and
potentially different message) that will verify correctly. In other words,
separation of the hybrid signature into component signatures implies that the
component signature will fail verification (of the component signature
scheme) entirely. Therefore, authentication is provided by the sender to the
receiver through correct verification of the digital signature(s), as in
traditional signature security experiments. It is not dependent on other
components, such as message content checking, or protocol level aspects, such
as public key provenance. As an illustrative example distinguishing WNS from
SNS, consider the case of component algorithms Sigma_1.Sign
and
Sigma_2.Sign
where the hybrid signature is computed as a concatenation
(sig_1, sig_2)
, where sig_1 = Sigma_1.Sign(hybridAlgID, m)
and sig_2 =
Sigma_2.Sign(hybridAlgID, m)
. In this case, a new message m' =
(hybridAlgID, m)
along with signature sig_1
and Sigma_1.pk
, with the
hybrid artifact embedded in the message instead of the signature, could be
correctly verified. The separation would be identifiable through further
investigation but the signature verification itself would not fail. Thus,
this case shows WNS (assuming the verification algorithm is defined
accordingly) but not SNS.¶
Some work [I-D.ounsworth-pq-composite-sigs] has looked at reliance on the
public key certificate chains to explicitly define hybrid use of the public
key. Namely, that Sigma_1.pk
cannot be used without Sigma_2.pk
. This
implies pushing the hybrid artifacts into the protocol and system level and a
dependency on the security of other verification algorithms (namely those in
the certificate chain). This further requires that security analysis of a
hybrid digital signature requires analysis of the key provenance, i.e., not
simply that a valid public key is used but how its hybridization and hybrid
artifacts have been managed throughout the entire chain. External
dependencies such as this may imply hybrid artifacts lie outside the scope of
the signature algorithm itself. SNS may potentially be achievable based on
dependencies at the system level.¶
Backwards compatibility refers to the property where a hybrid signature may be verified by only verifying one component signature, allowing the scheme to be used by legacy receivers. In general this means verifying the traditional component signature scheme, potentially ignoring the post-quantum signature entirely. This provides an option to transition sender systems to post-quantum algorithms while still supporting select legacy receivers. Notably, this is a verification property; the sender has provided a hybrid digital signature, but the verifier is allowed, due to internal policy and/or implementation, to only verify one component signature. Backwards compatibility may be further decomposed to subcategories where component key provenance is either separate or hybrid so as to support implementations that cannot recognize (and/or process) hybrid signatures or keys.¶
Forwards compatibility has also been a consideration in hybrid proposals [I-D.becker-guthrie-noncomposite-hybrid-auth]. Forward compatibility assumes that hybrid signature schemes will be used for some time, but that eventually all systems will transition to use only one (particularly, only one post-quantum) algorithm. As this is very similar to backwards compatibility, it also may imply separability of a hybrid algorithm; however, it could also simply imply capability to support separate component signatures. Thus the key distinction between backwards and forwards compatibility is that backwards compatibility may be needed for legacy systems that cannot use and/or process hybrid or post-quantum signatures, whereas in forwards compatibility the system has those capabilities and can choose what to support (e.g., for efficiency reasons).¶
As noted in [I-D.ietf-tls-hybrid-design], ideally, forward/backward compatibility is achieved using redundant information as little as possible.¶
Simultaneous Verification (SV) builds on SNS and was first introduced in [HYBRIDSIGDESIGN]. SV requires that not only are all component signatures needed to achieve a successful verification present in the hybrid signature, but also that verification of both component algorithms occurs simultaneously. Namely, "missing" information needs to be computed by the verifier so they cannot “quit” the verification process before both component signatures are verified. SV mimics traditional digital signatures guarantees, essentially ensuring that the hybrid digital signature behaves as a single algorithm vs. two separate component stages. Alternatively phrased, under an SV guarantee it is not possible for an unerring verifier to initiate termination of the hybrid verification upon successful verification of one component algorithm without also knowing if the other component succeeded or failed.¶
Hybrid generality means that a general signature combiner is defined, based on inherent and common structures of component digital signatures "categories." For instance, since multiple signature schemes use a Fiat-Shamir Transform, a hybrid scheme based on the transform can be made that is generalizable to all such signatures. Such generality can also result in simplified constructions whereas more tailored hybrid variants might be more efficient in terms of sizes and performance.¶
Similarly to performance goals noted for hybridization of other cryptographic components [I-D.ietf-tls-hybrid-design] hybrid signature constructions are expected to be as performant as possible. For most hybrid signatures this means that the computation time should only minimally exceed the sum of the component signature computation time. It is noted that performance of any variety may come at the cost of other properties, such as hybrid generality.¶
Similarly to space considerations in [I-D.ietf-tls-hybrid-design], hybrid signature constructions are expected to be as space performant as possible. This includes messages (as they might increase if artifacts are used), public keys, and the hybrid signature. For the hybrid signature, size should no more than minimally exceed the signature size of the two component signatures. In some cases, it may be possible for a hybrid signature to be smaller than the concatenation of the two component signatures.¶
Duplicated information should be avoided when possible, as a general point of efficiency. This might include repeated information in hybrid certificates or in the communication of component certificates in additional to hybrid certificates (for example to achieve backwards/forwards-compatibility), or sending multiple public keys or signatures of the same component algorithm.¶
Non-separability is not a singular definition but rather is a scale,
representing degrees
of separability hardness, visualized in
Figure 1.¶
At one end of the spectrum are schemes in which one of the component signatures can be stripped away with the verifier not being able to detect the change during verification. An example of this includes simple concatenation of signatures without any artifacts used. Nested signatures (where a message is signed by one component algorithm and then the message-signature combination is signed by the second component algorithm) may also fall into this category, dependent on whether the inner or outer signature is stripped off without any artifacts remaining.¶
Next on the spectrum are weakly non-separable signatures. Under Weak Non-Separability, if one of the component signatures of a hybrid is removed artifacts of the hybrid will remain (in the message, signature, or at the protocol level, etc.). This may enable the verifier to detect if a component signature is stripped away from a hybrid signature, but that detectability depends highly on the type of artifact and permissions. For instance, if a message contains a label artifact "This message must be signed with a hybrid signature" then the system must be allowed to analyze the message contents for possible artifacts. Whether a hybrid signature offers (Weak/Strong) Non-Separability might also depend on the implementation and policy of the protocol or application the hybrid signature is used in on the verifier side. Such policies may be further ambiguous to the sender, meaning that the type of authenticity offered to the receiver is unclear. In another example, under nested signatures the verifier could be tricked into interpreting a new message as the message/inner signature combination and verify only the outer signature. In this case, the inner signature-tag is an artifact.¶
Third on the scale is the Strong Non-Separability notion, in which separability detection is dependent on artifacts in the signature itself. Unlike in Weak Non-Separability, where artifacts may be in the actual message, the certificate, or in other non-signature components, this notion more closely ties to traditional algorithm security notions (such as EUF-CMA) where security is dependent on the internal construct of the signature algorithm and its verification. In this type, the verifier can detect artifacts on an algorithmic level during verification. For example, the signature itself may encode the information that a hybrid signature scheme is used. Examples of this type may be found in [HYBRIDSIGDESIGN].¶
For schemes achieving the most demanding security notion, Strong
Non-Separability with Simultaneous Verification, verification succeeds not
only when both of the component signatures are present but also only when the
verifier has verified both signatures. Moreover, no information is leaked to
the receiver during the verification process on the possible
validity/invalidity of the component signatures until both verify (or fail to
verify). This construct most closely mirrors traditional digital signatures
where, assuming that the verifier does verify a signature at all, the result
is either a positive verification of the full signature or a failure if the
signature is not valid. For fused hybrid signatures, a full signature
implies the fusion of both component algorithms, and therefore the strongest
non-separability notion enforces an all-or-nothing approach to
verification. Examples of algorithms providing this type of security can be
found in [HYBRIDSIGDESIGN].¶
Hybridization benefits from the presence of artifacts as evidence of the sender's intent to decrease the risk of successful stripping attacks. This, however, depends strongly on where such evidence resides (e.g., in the message, the signature, or somewhere on the protocol level instead of at the algorithmic level). Even commonly discussed hybrid approaches, such as concatenation, are not inherently tied to one type of security (e.g., WNS or SNS). This can lead to ambiguities when comparing different approaches and assumptions about security or lack thereof. Thus in this section we cover artifact locations and also walk through a high-level comparison of a few hybrid categories to show how artifact location can differ within a given approach. Artifact location is tied to non-separability notions above; thus the selection of a given security guarantee and general hybrid approach must also include finer grained selection of artifact placement.¶
There are a variety of artifact locations possible, ranging from within the message to the signature algorithm to the protocol level and even into policy, as shown in Table 1. For example, one artifact location could be in the message to be signed, e.g., containing a label artifact. Depending on the hybrid type, it might be possible to strip this away. For example, a quantum attacker could strip away the post-quantum signature of a concatenated dual signature, and (being able to forge, e.g., ECDSA signatures) remove the label artifact from the message as well. So, for many applications and threat models, adding an artifact in the message might be insufficient under stripping attacks. Another artifact location could be in the public key certificates as described in [I-D.ounsworth-pq-composite-sigs]. In such a case, the artifacts are still present even if a stripping attack occurs. In yet another case, artifacts may be present through the fused hybrid method, thus making them part of the signature at the algorithmic level. Note that in this latter case, it is not possible for an adversary to strip one of the component signatures or use a component of the hybrid to create a forgery for a component algorithm. Such signatures provide SNS. This consequently also implies that the artifacts of hybridization are absolute in that verification failure would occur if an adversary tries to remove them.¶
Eventual security analysis may be a consideration in choosing between levels. For example, if the security of the hybrid scheme is dependent on system policy, then cryptographic analysis must necessarily be reliant on specific policies and it may not be possible to describe a scheme's security in a standalone sense.¶
Location of artifacts of hybrid intent | Level |
---|---|
Signature | Algorithm |
Certificate | Protocol |
Algorithm agreement / negotiation | Protocol |
Message | Policy |
Here we provide a high-level example of how artifacts can appear in different locations even within a single, common approach. We look at the following categories of approaches: concatenation, nesting, and fusion. This is to illustrate that a given approach does not inherently imply a specific non-separability notion and that there are subtleties to the selection decision, since hybrid artifacts are related to non-separability guarantees. Additionally, this comparison highlights how artifacts placement can be identical in two different hybrid approaches.¶
We briefly summarize the hybrid approach categories (concatenation, nesting, and fusion) for clarity in description, before showing how each one may have artifacts in different locations in Table 2.¶
Concatenation: variants of hybridization where, for component algorithms
Sigma_1.Sign
and Sigma_2.Sign
, the hybrid signature is calculated as a
concatenation (sig_1, sig_2)
such that sig_1 = Sigma_1.Sign(hybridAlgID,
m)
and sig_2 = Sigma_2.Sign(hybridAlgID, m)
.¶
Nesting: variants of hybridization where for component algorithms
Sigma_1.Sign
and Sigma_2.Sign
, the hybrid signature is calculated in a
layered approach as (sig_1, sig_2)
such that, e.g., sig_1 =
Sigma_1.Sign(hybridAlgID, m)
and sig_2 = Sigma_2.Sign(hybridAlgID, (m,
sig_1))
.¶
Fused hybrid: variants of hybridization where for component algorithms
Sigma_1.Sign
and Sigma_2.Sign
, the hybrid signature is calculated to
generate a single hybrid signature sig_h
that cannot be cleanly separated
to form one or more valid component constructs. For example, if both
signature schemes are signatures schemes constructed through the Fiat-Shamir
transform, the component signatures would include responses r_1 and r_2 and
challenges c_1 and c_2, where c_1 and c_2 are hashes computed over the
respective commitments comm_1 and comm_2 (and the message). A fused hybrid
signature could consist of the component responses, r_1 and r_2 and a
challenge c that is computed as a hash over both commitments, i.e., c =
Hash(comm_1,comm_2,message). As such, c does not belong to either of the
component signatures but rather both, meaning that the signatures are
'entangled'.¶
# | Location of artifacts of hybrid intent | Category |
---|---|---|
Concatenated | ||
1 | None | No label in message, public keys are in separate certs |
2 | In message | Label in message, public keys are in separate certs |
3 | In cert | No label in message, public keys are in combined cert |
4 | In message and cert | Label in message, public keys are in combined cert |
Nested | ||
5 | In message | Label in message, public keys are in separate certs |
6 | In cert | No label in message, public keys are in combined cert |
7 | In message and cert | Label in message, public keys are in combined cert |
Fused | ||
8 | In signature | Public keys are in separate certs |
9 | In signature and message | Label in message, public keys are in separate certs |
10 | In signature and cert | Public keys are in combined cert |
11 | In signature and message and cert | Label in message, public keys are in combined cert |
As can be seen, while concatenation may appear to refer to a single type of combiner, there are in fact several possible artifact locations depending on implementation choices. Artifacts help to support detection in the case of stripping attacks, which means that different artifact locations imply different overall system implementation considerations to be able to achieve such detection.¶
Case 1 provides the weakest guarantees of hybrid identification, as there are no prescribed artifacts and therefore non-separability is not achieved. However, as can be seen, this does not imply that every implementation using concatenation fails to achieve non-separability. Thus, it is advisable for implementors to be transparent about artifact locations.¶
In cases 2 and 5 the artifacts lie within the message. This is notable as the authenticity of the message relies on the validity of the signature, and the artifact location means that the signature in turn relies on the authentic content of the message (the artifact label). This creates a risk of circular dependency. Alternative approaches such as cases 3 and 4 solve this circular dependency by provisioning keys in a combined certificate.¶
Another observation from this comparison is that artifact locations may be similar among some approaches. For instance, case 3 and case 6 both contain artifacts in the certificate. Naturally these examples are high-level and further specification on concrete schemes in the categories are needed before prescribing non-separability guarantees to each, but this does indicate how there could be a strong similarity between such guarantees. Such comparisons allow for a systematic decision process, where security is compared and identified and, if schemes are similar in the desired security goal, then decisions between schemes can be based on performance and implementation ease.¶
A final observation that this type of comparison provides is how various combiners may change the security analysis assumptions in a system. For instance, cases 3, 4, 5, and 6 all push artifacts - and therefore the signature validity - into the certificate chain. Naturally the entire chain must then also use a similar combiner if a straightforward security argument is to be made. Other cases, such as 8, 9, 10, and 11 put artifacts within the signature itself, meaning that these bear the closest resemblance to traditional schemes where message authenticity is dependent on signature validity.¶
In practice, use of hybrid digital signatures relies on standards specifications where applicable. This is particularly relevant in the case of FIPS approval considerations as well as NIST, which has provided basic guidance on hybrid signature use. NIST provides the following guidance (emphasis added),¶
Assume that in a [hybrid] signature, one signature is generated
with a NIST-approved signature scheme as specified in FIPS 186, while
another signature(s) can be generated using different schemes, e.g.,
ones that are not currently specified in NIST standards...hybrid
signatures
can be accommodated by current standards in FIPS mode,
as defined in FIPS 140, provided at least one of the component methods
is a properly implemented, NIST-approved signature algorithm. For the
purposes of FIPS 140 validation, any signature that is generated by a
non-approved component scheme would not be considered a security
function, since the NIST-approved component is regarded as assuring
the validity of the hybrid
signature. [NIST_PQC_FAQ]¶
The emphasized texts point to two things: 1) the signature scheme for
one of the component algorithms must be approved and 2) that said
algorithm must be properly implemented. This leaves some ambiguity as to
whether only the algorithm must be approved and well implemented, or if
that implementation must go through an approval process as well. As
such, there is a scale of approval
that developers may consider as
to whether they are using at least one approved component algorithm
(1-out-of-n approved software module
), or whether the implementation
of that component algorithm has gone through an approvals review (thus
making a all approved software module
). The former 1-out-of-n
approved software module
would suggest a straightforward path for
FIPS-140 approvals based on the NIST guidelines; however, it is not
inconceivable that using a all approved software module
could
automate much of the certification review and therefore be attractive to
developers.¶
We provide a scale for the different nuances of approval of the hybrid combiners. This is related to whether the combiner needs a new approval process or falls under already approved specifications.¶
The first listed "combiner" would be a new construction with a security reduction to different hardness assumptions but not necessarily to approved (or even existing) signature schemes. Such a new, singular algorithm relies on both traditional and nextgen principles.¶
Next, is a combiner that might take inspiration from existing/approved signature schemes such that its security can be reduced to the security of the approved algorithms. The combiner may, however, alter the implementations. As such it is uncertain whether new approval would be needed as it might depend on the combiner and changes. Such a case may potentially imply a distinction between a need for fresh approval of the algorithm(s) and approval of the implementation(s).¶
The 1-out-of-n combiner uses at least one approved algorithm implementation in a black-box way. It may potentially change the specifics of the other component algorithm implementations. As long as at least one component is approved, no new approval is needed (per [NIST_PQC_FAQ]).¶
In an All-Approved combiner, all algorithm implementations are used in a black-box way. A concatenation combiner is a simple example (where a signature is valid if all component signatures are valid). As long as at least one component is approved, no new approval is needed (per [NIST_PQC_FAQ]); thus as all algorithm implementations are approved the requirement is satisfied.¶
Under traditional signature scheme security assumptions such as EUF-CMA, the
adversary 'wins' the security experiment if it can produce a new message such
that a message-signature pair (m, sig)
correctly verifies. This traditional
security notion is challenged under a hybrid construct.¶
The most straightforward comparison would be for the adversary to attempt to
produce a new message m'
that a message-hybrid signature pair (m', sig_h)
correctly verifies. However, such a guarantee depends on the signature being
strongly non-separable. Otherwise, in practical terms a security experiment
must capture the case that an existing or new message m
could be verified
with a component signature, e.g., to produce (m', sig_1)
that correctly
verifies under Sigma_1.Sign
. Such considerations are beyond the scope of
traditional security analysis and represent considerations that would need to
be accounted for depending on the signature combiner method chosen.¶
This document discusses digital signature constructions that may be used in security protocols. It is an informational document and does not directly affect any other Internet draft. The security considerations for any specific implementation or incorporation of a hybrid scheme should be discussed in the relevant specification documents.¶
The design (and hence, security guarantees) of hybrid signature schemes depend heavily on the properties needed for the application or protocol using hybrid signatures. It seems that not all goals can be achieved simultaneously as exemplified below.¶
There is an inherent mutual exclusion between backwards compatibility and SNS. While WNS allows for a valid separation under leftover artifacts, SNS will ensure verification failure if a receiver attempts separation.¶
Similarly, there is an inherent mutual exclusion between backwards compatibility, when acted upon, and hybrid unforgeability as briefly mentioned above. Since the goal of backwards compatibility is usually to allow legacy systems without any software change to be able to process hybrid signatures, all differences between the legacy signature format and the hybrid signature format must be allowed to be ignored, including skipping verification of signatures additional to the classical signature. As such, if a system does skip an component signature, security does not rely on the security of all component signatures. Note that this mutual exclusion occurs at the verification stage, as a hybrid signature that is verified by a system that can process both component schemes can provide hybrid unforgeability even if another (legacy) system, processing the same hybrid signature, loses that property.¶
It seems that the more simultaneous verification is enforced by the hybrid design, the higher is the need-for-approval as simultaneous verification algorithms fuse (or 'entangle') the verification of the component algorithms such that verification operations from the different component schemes depend on each other in some way. For example, concatenation of signatures in a black-box way without any artefacts is, e.g., FIPS-approved, but the component signatures are usually verified separately and no 'simultaneous verification' is enforced.¶
This draft is based on the template of [I-D.ietf-tls-hybrid-design].¶
We would like to acknowledge the following people in alphabetical order who have contributed to pushing this draft forward, offered insights and perspectives, and/or stimulated work in the area:¶
Scott Fluhrer, Felix Günther, John Gray, Serge Mister, Max Pala, Mike Ounsworth, Douglas Stebila, Brendan Zember¶