This specification describes mechanisms for ensuring the authenticity and integrity of Verifiable Credentials and similar types of constrained digital documents using cryptography, especially through the use of digital signatures and related mathematical proofs.
The Working Group is actively seeking implementation feedback for this specification. In order to exit the Candidate Recommendation phase, the Working Group has set the requirement of at least two independent implementations for each mandatory feature in the specification. For details on the conformance testing process, see the test suites listed in the implementation report.
This specification describes mechanisms for ensuring the authenticity and integrity of Verifiable Credentials and similar types of constrained digital documents using cryptography, especially through the use of digital signatures and related mathematical proofs. Cryptographic proofs enable functionality that is useful to implementors of distributed systems. For example, proofs can be used to:
The operation of Data Integrity is conceptually simple. To create a cryptographic proof, the following steps are performed: 1) Transformation, 2) Hashing, and 3) Proof Generation.
Transformation is a process described by a transformation algorithm that takes input data and prepares it for the hashing process. One example of a possible transformation is to take a record of people's names that attended a meeting, sort the list alphabetically by the individual's family name, and rewrite the names on a piece of paper, one per line, in sorted order. Examples of transformations include canonicalization and binary-to-text encoding.
Hashing is a process described by a hashing algorithm that calculates an identifier for the transformed data using a cryptographic hash function. This process is conceptually similar to how a phone address book functions, where one takes a person's name (the input data) and maps that name to that individual's phone number (the hash). Examples of cryptographic hash functions include SHA-3 and BLAKE-3.
Proof Generation is a process described by a proof serialization algorithm that calculates a value that protects the integrity of the input data from modification or otherwise proves a certain desired threshold of trust. This process is conceptually similar to the way a wax seal can be used on an envelope containing a letter to establish trust in the sender and show that the letter has not been tampered with in transit. Examples of proof serialization functions include digital signatures and proofs of stake.
To verify a cryptographic proof, the following steps are performed: 1) Transformation, 2) Hashing, and 3) Proof Verification.
During verification, the [=transformation=] and [=hashing=] steps are conceptually the same as described above.
Proof Verification is a process that is described by a proof verification algorithm that applies a cryptographic proof verification function to see if the input data can be trusted. Possible proof verification functions include digital signatures and proofs of stake.
This specification details how cryptographic software architects and implementers can package these processes together into things called [=cryptographic suites=] and provide them to application developers for the purposes of protecting the integrity of application data in transit and at rest.
This specification optimizes for the following design goals:
While this specification primarily focuses on Verifiable Credentials, the design of this technology is generalized, such that it can be used for non-Verifiable Credential use cases. In these instances, implementers are expected to perform their own due diligence and expert review as to the applicability of the technology to their use case.
A conforming secured document is any [=byte sequence=] that can be converted to a JSON document that follows the relevant normative requirements in Sections [[[#proofs]]], [[[#proof-purposes]]], [[[#contexts-and-vocabularies]]], and [[[#dataintegrityproof]]].
A conforming cryptographic suite specification is any specification that follows the relevant normative requirements in Section [[[#cryptographic-suites]]].
A conforming processor is any algorithm realized as software and/or hardware that generates and/or consumes a [=conforming secured document=] according to the relevant normative statements in Section [[[#algorithms]]]. Conforming processors MUST produce errors when non-conforming documents are consumed.
This section defines the terms used in this specification. A link to these terms is included whenever they appear in this specification.
example.com
, an
ad-hoc value such as mycorp-level3-access
, or a very
specific transaction value like 8zF6T8J34qP3mqP
. A signer could
include a [=domain=] in its digital proof to restrict its use
to particular target, identified by the specified [=domain=].
A set of parameters that can be used together with a process to independently verify a proof. For example, a cryptographic public key can be used as a verification method with respect to a digital signature; in such usage, it verifies that the signer possessed the associated cryptographic private key.
"Verification" and "proof" in this definition are intended to apply broadly. For example, a cryptographic public key might be used during Diffie-Hellman key exchange to negotiate a shared symmetric key for encryption. This guarantees the integrity of the key agreement process. It is thus another type of verification method, even though descriptions of the process might not use the words "verification" or "proof."
An expression of the relationship between the [=subject=] and a [=verification method=]. An example of a verification relationship is authentication.
This section specifies the data model that is used for expressing [=data integrity proofs=], [=controller documents=], and [=verification methods=].
All of the data model properties and types in this specification map to URLs. The vocabulary where these URLs are defined is the [[?SECURITY-VOCABULARY]]. The explicit mechanism that is used to perform this mapping in a secured document is the `@context` property.
The mapping mechanism is defined by JSON-LD [[JSON-LD11]]. To ensure a document can be interoperably consumed without the use of a JSON-LD library, document authors are advised to ensure that domain experts have 1) specified the expected order for all values associated with a `@context` property, 2) published cryptographic hashes for each `@context` file, and 3) deemed that the contents of each `@context` file are appropriate for the intended use case.
When a document is processed by a non-JSON-LD processor and there is a requirement to use the same semantics as those used in a JSON-LD environment, implementers are advised to 1) enforce the expected order and values in the `@context` property, and 2) ensure that each `@context` file matches the known cryptographic hashes for each `@context` file.
Using static, versioned `@context` files with published cryptographic hashes in conjunction with JSON Schema is one acceptable approach to implementing the mechanisms described above, which ensures proper term identification, typing, and order, when a non-JSON-LD processor is used.
A [=data integrity proof=] provides information about the proof mechanism, parameters required to verify that proof, and the proof value itself. All of this information is provided using Linked Data vocabularies such as the [[?SECURITY-VOCABULARY]].
When expressing a [=data integrity proof=] on an object, a `proof` property MUST be used. The `proof` property within a Verifiable Credential is a named graph. If present, its value MUST be either a single object, or an unordered set of objects, expressed using the properties below:
A proof can be added to a JSON document like the following:
{ "myWebsite": "https://hello.world.example/" };
The following proof secures the document above using the `eddsa-jcs-2022` cryptography suite [[?DI-EDDSA]], which produces a verifiable digital proof by transforming the input data using the JSON Canonicalization Scheme (JCS) [[?RFC8785]] and then digitally signing it using an Edwards Digital Signature Algorithm (EdDSA).
{ "myWebsite": "https://hello.world.example/", "proof": { "type": "DataIntegrityProof", "cryptosuite": "eddsa-jcs-2022", "created": "2023-03-05T19:23:24Z", "verificationMethod": "https://di.example/issuer#z6MkjLrk3gKS2nnkeWcmcxiZPGskmesDpuwRBorgHxUXfxnG", "proofPurpose": "assertionMethod", "proofValue": "zQeVbY4oey5q2M3XKaxup3tmzN4DRFTLVqpLMweBrSxMY2xHX5XTYV 8nQApmEcqaqA3Q1gVHMrXFkXJeV6doDwLWx" } }
Similarly, a proof can be added to a JSON-LD data document like the following:
{ "@context": {"myWebsite": "https://vocabulary.example/myWebsite"}, "myWebsite": "https://hello.world.example/" };
The following proof secures the document above by using the `ecdsa-rdfc-2019` cryptography suite [[?DI-ECDSA]], which produces a verifiable digital proof by transforming the input data using the RDF Dataset Canonicalization Scheme [[?RDF-CANON]] and then digitally signing it using the Elliptic Curve Digital Signature Algorithm (ECDSA).
{ "@context": [ {"myWebsite": "https://vocabulary.example/myWebsite"}, "https://w3id.org/security/data-integrity/v2" ], "myWebsite": "https://hello.world.example/", "proof": { "type": "DataIntegrityProof", "cryptosuite": "ecdsa-rdfc-2019", "created": "2020-06-11T19:14:04Z", "verificationMethod": "https://ldi.example/issuer#zDnaepBuvsQ8cpsWrVKw8fbpGpvPeNSjVPTWoq6cRqaYzBKVP", "proofPurpose": "assertionMethod", "proofValue": "zXb23ZkdakfJNUhiTEdwyE598X7RLrkjnXEADLQZ7vZyUGXX8cyJZR BkNw813SGsJHWrcpo4Y8hRJ7adYn35Eetq" } }
This specification enables the expression of dates and times, such as through the `created` and `expires` properties. This information might be indirectly exposed to an individual if a proof is processed and is detected to be outside an allowable time range. When displaying date and time values related to the validity of cryptographic proofs, implementers are advised to respect the locale and local calendar preferences of the individual [[?LTLI]]. Conversion of timestamps to local time values are expected to consider the time zone expectations of the individual. See for more details about representing time values to individuals.
The Data Integrity specification supports the concept of multiple proofs in a single document. There are two types of multi-proof approaches that are identified: Proof Sets (un-ordered) and Proof Chains (ordered).
A proof set is useful when the same data needs to be secured by multiple entities, but where the order of proofs does not matter, such as in the case of a set of signatures on a contract. A proof set, which has no order, is represented by associating a set of proofs with the `proof` key in a document.
{ "@context": [ {"myWebsite": "https://vocabulary.example/myWebsite"}, "https://w3id.org/security/data-integrity/v2" ], "myWebsite": "https://hello.world.example/", "proof": [{ "type": "DataIntegrityProof", "cryptosuite": "eddsa-rdfc-2022", "created": "2020-11-05T19:23:24Z", "verificationMethod": "https://ldi.example/issuer/1#z6MkjLrk3gKS2nnkeWcmcxiZPGskmesDpuwRBorgHxUXfxnG", "proofPurpose": "assertionMethod", "proofValue": "z4oey5q2M3XKaxup3tmzN4DRFTLVqpLMweBrSxMY2xHX5XTYVQeVbY8 nQAVHMrXFkXJpmEcqdoDwLWxaqA3Q1geV6" }, { "type": "DataIntegrityProof", "cryptosuite": "eddsa-rdfc-2022", "created": "2020-11-05T13:08:49Z", "verificationMethod": "https://pfps.example/issuer/2#z6MkGskxnGjLrk3gK S2mesDpuwRBokeWcmrgHxUXfnncxiZP", "proofPurpose": "assertionMethod", "proofValue": "z5QLBrp19KiWXerb8ByPnAZ9wujVFN8PDsxxXeMoyvDqhZ6Qnzr5CG9 876zNht8BpStWi8H2Mi7XCY3inbLrZrm95" }] }
A proof chain is useful when the same data needs to be signed by multiple entities and the order of when the proofs occurred matters, such as in the case of a notary counter-signing a proof that had been created on a document. A proof chain, where proof order needs to be preserved, is expressed by providing at least one proof with an `id`, such as a UUID as a URN, and another proof with a `previousProof` value that identifies the previous proof.
{ "@context": [ {"myWebsite": "https://vocabulary.example/myWebsite"}, "https://w3id.org/security/data-integrity/v2" ], "myWebsite": "https://hello.world.example/", "proof": [{ "id": "urn:uuid:60102d04-b51e-11ed-acfe-2fcd717666a7", "type": "DataIntegrityProof", "cryptosuite": "eddsa-rdfc-2022", "created": "2020-11-05T19:23:42Z", "verificationMethod": "https://ldi.example/issuer/1#z6MkjLrk3gKS2nnkeWcmcxiZPGskmesDpuwRBorgHxUXfxnG", "proofPurpose": "assertionMethod", "proofValue": "zVbY8nQAVHMrXFkXJpmEcqdoDwLWxaqA3Q1geV64oey5q2M3XKaxup 3tmzN4DRFTLVqpLMweBrSxMY2xHX5XTYVQe" }, { "type": "DataIntegrityProof", "cryptosuite": "eddsa-rdfc-2022", "created": "2020-11-05T21:28:14Z", "verificationMethod": "https://pfps.example/issuer/2#z6MkGskxnGjLrk3g KS2mesDpuwRBokeWcmrgHxUXfnncxiZP", "proofPurpose": "assertionMethod", "proofValue": "z6Qnzr5CG9876zNht8BpStWi8H2Mi7XCY3inbLrZrm955QLBrp19Ki WXerb8ByPnAZ9wujVFN8PDsxxXeMoyvDqhZ", "previousProof": "urn:uuid:60102d04-b51e-11ed-acfe-2fcd717666a7" }] }
When securing data in a document, it is important to clearly delineate the data being protected, which is every graph expressed in the document except the one containing the data associated with a securing mechanism, which is called a proof graph. Creating this separation enables the processing algorithms to deterministically protect and verify a secured document.
The information contained in an unsecured document before a [=data integrity proof=] is added to the document is expressed in one or more graphs. To ensure that information from different [=data integrity proofs=] is not accidentally co-mingled, the concept of a [=proof graph=] is used to encapsulate each [=data integrity proof=]. Each value associated with the `proof` property of the document identifies a separate graph, which is sometimes referred to as a named graph, of type ProofGraph, which contains a single [=data integrity proof=].
Using these graphs has a concrete effect when performing JSON-LD processing, as this properly separates statements expressed in one graph from those in another graph. Implementers that limit their processing to other media types, such as JSON, YAML, or CBOR, will need to keep this in mind if they merge data from one document with data from another, such as when an `id` value string is the same in both documents. It is important to not merge objects that seem to have similar properties, when those objects do not have an `id` property and/or use a global identifier type such as a URL, as without these, is not possible to tell whether two such objects are expressing information about the same entity.
A proof that describes its purpose helps prevent it from being misused for some other purpose.
Add a mention of JWK's `key_ops` parameter and WebCrypto's
`KeyUsage` restrictions; explain that Proof Purpose serves a
different goal and allows for finer-grained restrictions.
Dave Longley suggested that proof purposes enable verifiers to know what the
proof creator's intent was so the message can't be accidentally abused for
another purpose, e.g., a message signed for the purpose of merely making an
assertion (and thus perhaps intended to be widely shared) being abused as a
message to authenticate to a service or take some action (invoke a capability).
It's a goal to keep the number of them limited to as few categories as are
really needed to accomplish this goal.
The following is a list of commonly used [=proof purpose=] values.
Note: The Authorization Capabilities [[?ZCAP]] specification defines additional proof purposes for that use case, such as `capabilityInvocation` and `capabilityDelegation`.
When a link to an external resource is included in a [=conforming secured document=], it is desirable to know whether the resource that is identified has changed since the proof was created. This applies to cases where there is an external resource that is remotely retrieved as well as to cases where the [=verifier=] might have a locally cached copy of the resource.
To enable confirmation that a resource referenced by a [=conforming secured document=] has not changed since the document was secured, an implementer MAY include a property named `digestMultibase` in any object that includes an `id` property. If present, the `digestMultibase` value MUST be a single string value, or an array of string values, each of which is a Multibase-encoded Multihash value.
JSON-LD context authors are expected to add `digestMultibase` to contexts that will be used in documents that refer to other resources and to include an associated cryptographic digest. For example, the [[[VC-DATA-MODEL-2.0]]] refers to context (`https://www.w3.org/ns/credentials/v2`) which includes `digestMultibase`, and the [[[VC-DATA-MODEL-2.0]]] includes the hexadecimal encoded SHA2-256 digest value of that context document.
An example of a resource integrity protected object is shown below:
{ ... "image": { "id": "https://university.example.org/images/58473", "digestMultibase": "zQmdfTbBqBPQ7VNxZEYEj14VmRuZBkqFbiwReogJgS1zR1n" }, ... }
Implementers are urged to consult appropriate sources, such as the FIPS 180-4 Secure Hash Standard and the Commercial National Security Algorithm Suite 2.0 to ensure that they are choosing a hash algorithm that is appropriate for their use case.
The term Linked Data is used to describe a recommended best practice for exposing, sharing, and connecting information on the Web using standards, such as URLs, to identify things and their properties. When information is presented as Linked Data, other related information can be easily discovered and new information can be easily linked to it. Linked Data is extensible in a decentralized way, greatly reducing barriers to large scale integration.
With the increase in usage of Linked Data for a variety of applications, there is a need to be able to verify the authenticity and integrity of Linked Data documents. This specification adds authentication and integrity protection to data documents through the use of mathematical proofs without sacrificing Linked Data features such as extensibility and composability.
While this specification provides mechanisms to digitally sign Linked Data, the use of Linked Data is not necessary to gain some of the advantages provided by this specification.
Cryptographic suites that implement this specification can be used to secure [=verifiable credentials=] and [=verifiable presentations=]. Implementers that are addressing those use cases are cautioned that additional checks might be appropriate when processing those types of documents.
There are some use cases where it is important to ensure that the [=verification method=] used in a proof is associated with the `issuer` in a verifiable credential, or the `holder` in a verifiable presentation, during the process of validation. One way to check for such an association is to ensure that the value of the `controller` property of a proof's [=verification method=] matches the URL value used to identify the `issuer` or `holder`, respectively, and that the verification method is expressed under a verification relationship that is acceptable given the proof's purpose. This particular association indicates that the `issuer` or `holder`, respectively, is the controller of the [=verification method=] used to verify the proof.
Document authors and implementers are advised to understand the difference between the validity period of a proof, which is expressed using the `created` and `expires` properties, and the validity period of a credential, which is expressed using the `validFrom` and `validUntil` properties. While these properties might sometimes express the same validity periods, at other times they might not be aligned. When verifying a proof, it is important to ensure that the time of interest (which might be the current time or any other time) is within the validity period for the proof (that is, between `created` and `expires` ). When validating a [=verifiable credential=], it is important to ensure that the time of interest is within the validity period for the credential (that is, betweeen `validFrom` and `validUntil`). Note that a failure to validate either the validity period for the proof, or the validity period for the credential, might result in accepting data that ought to have been rejected.
Finally, implementers are also urged to understand that there is a difference between the revocation information associated with a [=verifiable credential=], and the revocation and expiration times for a [=verification method=]. The revocation and expiration times for a [=verification method=] are expressed using the `revocation` and `expires` properties, respectively; are related to events such as a private key being compromised or expiring; and can provide timing information which might reveal details about a controller, such as their security practices or when they might have been compromised. The revocation information for a [=verifiable credential=] is expressed using the `credentialStatus` property; is related to events such as an individual losing the privilege that is granted by the [=verifiable credential=]; and does not provide timing information, which enhances privacy.
This section lists cryptographic hash values that might change during the Candidate Recommendation phase based on implementer feedback that requires the referenced files to be modified.
Implementations that perform JSON-LD processing MUST treat the following JSON-LD context URLs as already resolved, where the resolved document matches the corresponding hash values below:
URL | Hash |
---|---|
https://w3id.org/security/data-integrity/v2 | sha256: `v/POI0jhSjPansxhJAP1fwepCBZ2HK77fRZfCCyBDs0=` |
https://w3id.org/security/multikey/v1 | sha256: `uiwYLeLZL35HGEvMqPzwvq7m05hsUnv2ZMGVu8fFhZc=` |
https://w3id.org/security/jwk/v1 | sha256: `9h/GLRVuGCl0rn/ye6lzf219aN7Sgzq9yBFqUoOI54k=` |
It is possible to confirm the cryptographic digests listed above by running a command like the following (replacing `<DOCUMENT_URL>` with the appropriate value) through a modern UNIX-like OS command line interface: `curl -sL -H "Accept: application/ld+json" <DOCUMENT_URL> | openssl dgst -sha256 -binary | openssl base64 -nopad -a`.
The security vocabulary terms that the JSON-LD contexts listed above resolve to are in the https://w3id.org/security# namespace. That is, all security terms in this vocabulary are of the form `https://w3id.org/security#TERM`, where `TERM` is the name of a term.
Implementations that perform RDF processing MUST treat the JSON-LD serialization of the vocabulary URL as already dereferenced, where the dereferenced document matches the corresponding hash value below.
When dereferencing the https://w3id.org/security# URL, the media type of the data that is returned depends on HTTP content negotiation. These are as follows:
Media Type | Description and Hash |
---|---|
application/ld+json |
The vocabulary in JSON-LD format [[?JSON-LD11]]. sha256: `LEaoTyf796eTaSlYWjfPe3Yb+poCW9TjWYTbFDmC0tc=` |
text/turtle |
The vocabulary in Turtle format [[?TURTLE]]. sha256: `McnhLyt7+/A/0iLb3CUXD0itNw+7bwwjtzOww/zwoyI=` |
text/html |
The vocabulary in HTML+RDFa Format [[?HTML-RDFA]]. sha256: `eUHP1xiSC157iTPDydZmxg/hvmX3g/nnCn+FO25d4dc=` |
It is possible to confirm the cryptographic digests listed above by running a command like the following (replacing `<MEDIA_TYPE>` and `<DOCUMENT_URL>` with the appropriate values) through a modern UNIX-like OS command line interface: `curl -sL -H "Accept: <MEDIA_TYPE>" <DOCUMENT_URL> | openssl dgst -sha256 -binary | openssl base64 -nopad -a`.
Authors of application-specific vocabularies and specifications SHOULD ensure that their JSON-LD context and vocabulary files are permanently cacheable using the approaches to caching described above or a functionally equivalent mechanism.
Implementations MAY load application-specific JSON-LD context files from the network during development, but SHOULD permanently cache JSON-LD context files used by [=conforming secured documents=] in production settings, to increase their security and privacy characteristics. Goals of processing speed MAY be achieved through caching approaches such as those described above or functionally equivalent mechanisms.
Some applications, such as digital wallets, that are capable of holding arbitrary [=verifiable credentials=] or other data-integrity-protected documents, from any issuer and using any contexts, might need to be able to load externally linked resources, such as JSON-LD context files, in production settings. This is expected to increase user choice, scalability, and decentralized upgrades in the ecosystem over time. Authors of such applications are advised to read the security and privacy sections of this document for further considerations.
For further information regarding processing of JSON-LD contexts and vocabularies, see Verifiable Credentials v2.0: Base Context and Verifiable Credentials v2.0: Vocabularies.
The `@context` property is used to ensure that implementations are using the same semantics when terms in this specification are processed. For example, this can be important when properties like `type` are processed and its value, such as `DataIntegrityProof`, are used.
When securing a document, if an `@context` property is not provided in the document or the Data Integrity terms used in the document are not mapped by existing values in the `@context` property, implementations MUST inject or add an `@context` property with a value of `https://w3id.org/security/data-integrity/v2`.
Context injection is expected to be unnecessary sometimes, such as when the Verifiable Credential Data Model v2.0 context (`https://www.w3.org/ns/credentials/v2`) exists as a value in the `@context` property, as that context maps all of the necessary Data Integrity terms that were previously mapped by `https://w3id.org/security/data-integrity/v2`.
HTML processors are designed to continue processing if recoverable errors are detected. JSON-LD processors operate in a similar manner. This design philosophy was meant to ensure that developers could use only the parts of the JSON-LD language that they find useful, without causing the processor to throw errors on things that might not be important to the developer. Among other effects, this philosophy led to JSON-LD processors being designed to not throw errors, but rather warn developers, when encountering things such as undefined terms.
When converting from JSON-LD to an RDF Dataset, such as when canonicalizing a document [[?RDF-CANON]], undefined terms and relative URLs can be dropped silently. When values are dropped, they are not protected by a digital proof. This creates a mismatch of expectations, where a developer, who is unaware of how a JSON-LD processor works, might think that certain data was being secured, and then be surprised to find that it was not, when no error was thrown. This specification requires that any recoverable loss of data when performing JSON-LD transformations result in an error, to avoid a mismatch in the security expectations of developers.
Implementations that use JSON-LD processing, such as RDF Dataset Canonicalization [[?RDF-CANON]], MUST throw an error, which SHOULD be `DATA_LOSS_DETECTION_ERROR`, when data is dropped by a JSON-LD processor, such as when an undefined term is detected in an input document.
Similarly, since [=conforming secured documents=] can be transferred from one security domain to another, [=conforming processors=] that process the [=conforming secured document=] cannot assume any particular base URL for the document. When deserializing to RDF, implementations MUST ensure that the base URL is set to null.
This section defines datatypes that are used by this specification.
This specification encodes cryptographic suite identifiers as enumerable strings, which is useful in processes that need to efficiently encode such strings, such as compression algorithms. In environments that support data types for string values, such as RDF [[?RDF-CONCEPTS]], cryptographic identifier content is indicated using a literal value whose datatype is set to `https://w3id.org/security#cryptosuiteString`.
The `cryptosuiteString` datatype is defined as follows:
Multibase-encoded strings are used to encode binary data into ASCII-only formats, which are useful in environments that cannot directly represent binary values. This specification makes use of this encoding. In environments that support data types for string values, such as RDF [[?RDF-CONCEPTS]], Multibase-encoded content is indicated using a literal value whose datatype is set to `https://w3id.org/security#multibase`.
The `multibase` datatype is defined as follows:
A data integrity proof is designed to be easy to use by developers and therefore strives to minimize the amount of information one has to remember to generate a proof. Often, just the [=cryptographic suite=] name (e.g. `eddsa-rdfc-2022`) is required from developers to initiate the creation of a proof. These [=cryptographic suite=]s are often created or reviewed by people that have the requisite cryptographic training to ensure that safe combinations of cryptographic primitives are used. This section specifies the requirements for authoring cryptographic suite specifications.
The requirements for all data integrity cryptographic suite specifications are as follows:
A [=cryptosuite instance=] is instantiated using a [=cryptosuite instantiation algorithm=] and is made available to algorithms in an implementation-specific manner. Implementations MAY use [[[VC-SPECS]]] [[VC-SPECS]] to discover known [=cryptosuite instantiation algorithms=].
The following language was deemed to be contentious: The specification MUST
provide a link to an interoperability test report to document which
implementations are conformant with the cryptographic suite specification.
The Working Group is seeking feedback on whether or not this is desired given
the important role that cryptographic suite specifications play in ensuring
data integrity.
A number of cryptographic suites follow the same basic pattern when expressing a data integrity proof. This section specifies that general design pattern, a cryptographic suite type called a `DataIntegrityProof`, which reduces the burden of writing and implementing cryptographic suites through the reuse of design primitives and source code.
When specifing a cryptographic suite that utilizes this design pattern, the `proof` value takes the following form:
Cryptographic suite designers MUST use mandatory `proof` value properties defined in Section [[[#proofs]]], and MAY define other properties specific to their cryptographic suite.
One of the design patterns seen in Data Integrity cryptosuites from
2012 to 2020 was use of the `type` property to establish a specific type for a
cryptographic suite. For example, the
Ed25519Signature2020 cryptographic suite was one such specification. This
led to a greater burden on cryptographic suite implementations, where every new
cryptographic suite required a new JSON-LD Context to be specified, resulting
in a sub-optimal developer experience. A streamlined version of this design
pattern emerged in 2020, such that a developer would only need to include a
single JSON-LD Context to support all modern cryptographic suites. This
encouraged more modern cryptosuites — such as the EdDSA Cryptosuites
[[?DI-EDDSA]] and the ECDSA Cryptosuites [[?DI-ECDSA]] — to be built
based on the streamlined pattern described in this section.
To improve the developer experience, authors creating new Data Integrity
cryptographic suite specifications SHOULD use the modern pattern — where
the `type` is set to `DataIntegrityProof`; the `cryptosuite` property carries
the identifier for the cryptosuite; and any cryptosuite-specific cryptographic
data is encapsulated (i.e., not directly exposed as application layer data)
within `proofValue`. A list of cryptographic suite specifications that are
known to follow this pattern is provided in the
Proof types section of
the Verifiable Credentials Specifications Directory.
The algorithms defined below operate on documents represented as JSON objects. This specification follows [[JSON-LD-API]] in representing a JSON object as an [[Infra]] [=map=]. An unsecured data document is a [=map=] with no `proof` [=map/key=]. A secured data document is a [=map=] with a `proof` [=map/key=].
Implementers MAY implement reasonable defaults and safeguards in addition to the algorithms below, to help mitigate developer error, excessive resource consumption, newly discovered attack models against which there is a particular protection, etc. The algorithms provided below are the minimum requirements for an interoperable implementation, and developers are urged to include additional measures that could contribute to a safer and more efficient ecosystem.
The following algorithm specifies how a digital proof can be added to an [=unsecured data document=], and can then be used to verify the output document's authenticity and integrity. Required inputs are an unsecured data document ([=map=] |unsecuredDocument|), a [=cryptosuite instance=] ([=struct=] |cryptosuite|), and a set of options ([=map=] |options|). Output is a [=secured data document=] ([=map=]) or an error. Whenever this algorithm encodes strings, it MUST use UTF-8 encoding.
The following algorithm specifies how to incrementally add a proof to a proof set or proof chain starting with a secured document containing either a proof or proof set/chain. Required inputs are a [=secured data document=] ([=map=] |securedDocument|), a [=cryptographic suite=] (|suite:cryptosuite instance|), and a set of options ([=map=] |options|). Output is a new [=secured data document=] ([=map=]). Whenever this algorithm encodes strings, it MUST use UTF-8 encoding.
This step is critical, as it binds
the previous proofs to the document
prior to signing. The |proof| value for the document will be updated
in a later step of this algorithm.
The following algorithm specifies how to check the authenticity and integrity of a [=secured data document=] by verifying its digital proof. The algorithm takes as input:
This algorithm returns a verification result, a [=struct=] whose [=struct/items=] are:
When a step says "an error MUST be raised", it means that a [=verification result=] MUST be returned with a [=verification result/verified=] of `false` and a non-empty [=verification result/errors=] list.
In a proof set or proof chain, a [=secured data document=] has a `proof` attribute which contains a list of proofs (|allProofs|). The following algorithm provides one method of checking the authenticity and integrity of a [=secured data document=], achieved by verifying every proof in |allProofs|. Other approaches are possible, particularly if it is only desired to verify a subset of the proofs contained in |allProofs|. If another approach is taken to verify only a subset of the proofs, then it is important to note that any proof in that subset with a `previousProof` can only be considered verified if the proofs it references are also considered verified.
Required input is a [=secured data document=] (|securedDocument|). A list of [=verification results=] corresponding to each proof in |allProofs| is generated, and a single combined [=verification result=] is returned as output. Implementations MAY return any of the other [=verification result=]s and/or any other metadata alongside the combined [=verification result=].
The algorithms described in this specification, as well as in various cryptographic suite specifications, throw specific types of errors. Implementers might find it useful to convey these errors to other libraries or software systems. This section provides specific URLs, descriptions, and error codes for the errors, such that an ecosystem implementing technologies described by this specification might interoperate more effectively when errors occur.
When exposing these errors through an HTTP interface, implementers SHOULD use [[RFC9457]] to encode the error data structure. If [[RFC9457]] is used:
The following section describes security considerations that developers implementing this specification should be aware of in order to create secure software.
Cryptography secures information through the use of secrets. Knowledge of the necessary secret makes it computationally easy to access certain information. The same information can be accessed if a computationally-difficult, brute-force effort successfully guesses the secret. All modern cryptography requires the computationally difficult approach to remain difficult throughout time, which does not always hold due to breakthroughs in science and mathematics. That is to say that Cryptography has a shelf life.
This specification plans for the obsolescence of all cryptographic approaches by asserting that whatever cryptography is in use today is highly likely to be broken over time. Software systems have to be able to change the cryptography in use over time in order to continue to secure information. Such changes might involve increasing required secret sizes or modifications to the cryptographic primitives used. However, some combinations of cryptographic parameters might actually reduce security. Given these assumptions, systems need to be able to distinguish different combinations of safe cryptographic parameters, also known as cryptographic suites, from one another. When identifying or versioning cryptographic suites, there are several approaches that can be taken which include: parameters, numbers, and dates.
Parametric versioning specifies the particular cryptographic parameters that are employed in a cryptographic suite. For example, one could use an identifier such as `RSASSA-PKCS1-v1_5-SHA1`. The benefit to this scheme is that a well-trained cryptographer will be able to determine all of the parameters in play by the identifier. The drawback to this scheme is that most of the population that uses these sorts of identifiers are not well trained and thus will not understand that the previously mentioned identifier is a cryptographic suite that is no longer safe to use. Additionally, this lack of knowledge might lead software developers to generalize the parsing of cryptographic suite identifiers such that any combination of cryptographic primitives becomes acceptable, resulting in reduced security. Ideally, cryptographic suites are implemented in software as specific, acceptable profiles of cryptographic parameters instead.
Numbered versioning might specify a major and minor version number such as `1.0` or `2.1`. Numbered versioning conveys a specific order and suggests that higher version numbers are more capable than lower version numbers. The benefit of this approach is that it removes complex parameters that less expert developers might not understand with a simpler model that conveys that an upgrade might be appropriate. The drawback of this approach is that its not clear if an upgrade is necessary, as software version number increases often don't require an upgrade for the software to continue functioning. This can lead to developers thinking their usage of a particular version is safe, when it is not. Ideally, additional signals would be given to developers that use cryptographic suites in their software that periodic reviews of those suites for continued security are required.
Date-based versioning specifies a particular release date for a specific cryptographic suite. The benefit of a date, such as a year, is that it is immediately clear to a developer if the date is relatively old or new. Seeing an old date might prompt the developer to go searching for a newer cryptographic suite, where as a parametric or number-based versioning scheme might not. The downside of a date-based version is that some cryptographic suites might not expire for 5-10 years, prompting the developer to go searching for a newer cryptographic suite only to not find one that is newer. While this might be an inconvenience, it is one that results in safer ecosystem behavior.
Modern cryptographic algorithms provide a number of tunable parameters and options to ensure that the algorithms can meet the varied requirements of different use cases. For example, embedded systems have limited processing and memory environments and might not have the resources to generate the strongest digital signatures for a given algorithm. Other environments, like financial trading systems, might only need to protect data for a day while the trade is occurring, while other environments might need to protect data for multiple decades. To meet these needs, cryptographic algorithm designers often provide multiple ways to configure a cryptographic algorithm.
Cryptographic library implementers often take the specifications created by cryptographic algorithm designers and specification authors and implement them such that all options are available to the application developers that use their libraries. This can be due to not knowing which combination of features a particular application developer might need for a given cryptographic deployment. All options are often exposed to application developers.
Application developers that use cryptographic libraries often do not have the requisite cryptographic expertise and knowledge necessary to appropriately select cryptographic parameters and options for a given application. This lack of expertise can lead to an inappropriate selection of cryptographic parameters and options for a particular application.
This specification sets the priority of constituencies to protect application developers over cryptographic library implementers over cryptographic specification authors over cryptographic algorithm designers. Given these priorities, the following recommendations are made:
The guidance above is meant to ensure that useful cryptographic options and parameters are provided at the lower layers of the architecture while not exposing those options and parameters to application developers who may not fully understand the balancing benefits and drawbacks of each option.
The VCWG is seeking guidance on adding language to allow the use of experimental or deprecated cryptography. By default, those features will be disabled and will require the application developer to specifically allow use on a per-cryptographic suite basis. There will be requirements for all implementing libraries to throw errors or warnings when deprecated or experimental options are selected without the appropriate override flags.
Section [[[#versioning-cryptography-suites]]] emphasized the importance of providing relatively easy to understand information concerning the timeliness of particular cryptographic suite, while section [[[#protecting-application-developers]]] further emphasized minimizing the number of options to be specified. Indeed, section [[[#cryptographic-suites]]] lists requirements for cryptographic suites which include detailed specification of algorithm, transformation, hashing, and serialization. Hence, the name of the cryptographic suite does not need to include all this detail, which implies the parametric versioning mentioned in section [[[#versioning-cryptography-suites]]] is neither necessary nor desirable.
The recommended naming convention for cryptographic suites is a string composed of a signature algorithm identifier, separated by a hyphen from an option identifier (if the cryptosuite supports incompatible implementation options), followed by a hyphen and designation of the approximate year that the suite was proposed.
For example, the [[?DI-EDDSA]] is based on EdDSA digital signatures, supports
two incompatible options based on canonicalization approaches, and was proposed
in roughly the year 2022, so it would have two different cryptosuite names:
eddsa-rdfc-2022
and eddsa-jcs-2022
.
Although the [[?DI-ECDSA]] is based on ECDSA digital signatures, supports the
same two incompatible canonicalization approaches as [[?DI-EDDSA]], and supports
two different levels of security (128 bit and 192 bit) via two alternative sets
of elliptic curves and hashes, it has only two cryptosuite names:
ecdsa-rdfc-2019
and ecdsa-jcs-2019
. The security level and corresponding
curves and hashes are determined from the multi-key format of the public key
used in validation.
Cryptographic agility is a practice by which one designs frequently connected information security systems to support switching between multiple cryptographic primitives and/or algorithms. The primary goal of cryptographic agility is to enable systems to rapidly adapt to new cryptographic primitives and algorithms without making disruptive changes to the systems' infrastructure. Thus, when a particular cryptographic primitive, such as the SHA-1 algorithm, is determined to be no longer safe to use, systems can be reconfigured to use a newer primitive via a simple configuration file change.
Cryptographic agility is most effective when the client and the server in the information security system are in regular contact. However, when the messages protected by a particular cryptographic algorithm are long-lived, as with Verifiable Credentials, and/or when the client (holder) might not be able to easily recontact the server (issuer), then cryptographic agility does not provide the desired protections.
Cryptographic layering is a practice where one designs rarely connected information security systems to employ multiple primitives and/or algorithms at the same time. The primary goal of cryptographic layering is to enable systems to survive the failure or one or more cryptographic algorithms or primitives without losing cryptographic protection on the payload. For example, digitally signing a single piece of information using RSA, ECDSA, and Falcon algorithms in parallel would provide a mechanism that could survive the failure of two of these three digital signature algorithms. When a particular cryptographic protection is compromised, such as an RSA digital signature using 768-bit keys, systems can still utilize the non-compromised cryptographic protections to continue to protect the information. Developers are urged to take advantage of this feature for all signed content that might need to be protected for a year or longer.
This specification provides for both forms of agility. It provides for cryptographic agility, which allows one to easily switch from one algorithm to another. It also provides for cryptographic layering, which allows one to simultaneously use multiple cryptographic algorithms, typically in parallel, such that any of those used to protect information can be used without reliance on or requirement of the others, while still keeping the digital proof format easy to use for developers.
At times, it is beneficial to transform the data being protected during the cryptographic protection process. Such "in-line" transformation can enable a particular type of cryptographic protection to be agnostic to the data format it is carried in. For example, some Data Integrity cryptographic suites utilize RDF Dataset Canonicalization [[?RDF-CANON]] which transforms the initial representation into a canonical form [[?N-QUADS]] that is then serialized, hashed, and digitally signed. As long as any syntax expressing the protected data can be transformed into this canonical form, the digital signature can be verified. This enables the same digital signature over the information to be expressed in JSON, CBOR, YAML, and other compatible syntaxes without having to create a cryptographic proof for every syntax.
Being able to express the same digital signature across a variety of syntaxes is beneficial because systems often have native data formats with which they operate. For example, some systems are written against JSON data, while others are written against CBOR data. Without transformation, systems that process their data internally as CBOR are required to store the digitally signed data structures as JSON (or vice-versa). This leads to double-storing data and can lead to increased security attack surface if the unsigned representation stored in databases accidentally deviates from the signed representation. By using transformations, the digital proof can live in the native data format to help prevent otherwise undetectable database drift over time.
This specification is designed to avoid requiring the duplication of signed information by utilizing "in-line" data transformations. Application developers are urged to work with cryptographically protected data in the native data format for their application and not separate storage of cryptographic proofs from the data being protected. Developers are also urged to regularly confirm that the cryptographically protected data has not been tampered with as it is written to and read from application storage.
Some transformations, such as RDF Dataset Canonicalization [[?RDF-CANON]], have mitigations for input data sets that can be used by attackers to consume excessive processing cycles. This class of attack is called dataset poisoning, and all modern RDF Dataset canonicalizers are required to detect these sorts of bad inputs and halt processing. The test suites for RDF Dataset Canonicalization includes such poisoned datasets to ensure that such mitigations exist in all conforming implementations. Generally speaking, cryptographic suite specifications that use transformations are required to mitigate these sorts of attacks, and implementers are urged to ensure that the software libraries that they use enforce these mitigations. These attacks are in the same general category as any resource starvation attack, such as HTTP clients that deliberately slow connections, thus starving connections on the server. Implementers are advised to consider these sorts of attacks when implementing defensive security strategies.
The VCWG is seeking feedback on normative language that cryptographic suite implementers need to follow to ensure that they do not utilize data transformation mechanisms that can map to the same output. That is, given different inputs for canonicalization scheme #1 and canonicalization scheme #2, they must not produce the same output value. As an analogy, this is the same requirement for cryptographic hashing mechanisms and is why those schemes are designed to be collision resistant. Cryptographic canonicalization mechanisms have the same requirement. At present, this isn't a problem because the three expected canonicalization schemes — the Universal RDF Dataset Canonicalization Algorithm 2015 [[?RDF-CANON]], JSON Canonicalization Scheme [[?RFC8785]], and a theoretical future base-encoding canonicalization — have entirely different outputs.
The VCWG is seeking feedback on whether to explain why modern canonicalization schemes are simpler than the far more complex XML Canonicalization schemes of the early 2000s. Some readers seem to be under the impression that all canonicalization is difficult and has to be avoided at all costs (including costs to application developers). The WG would like to understand if it would be helpful to include a section explaining why some simpler data syntaxes (such as JSON) are easier to canonicalize than more complex data syntaxes (such as XML).
The data that is protected by any [=data integrity proof=] is the [=transformation|transformed data=]. [=transformation|Transformed data=] is generated by a [=transformation algorithm=] that is specified by a particular [=cryptosuite=]. This protection mechanism differs from some more traditional digital signature mechanisms that do not perform any sort of [=transformation=] on the input data. The benefits of [=transformation=] are detailed in Section [[[#transformations]]].
For example, [=cryptosuites=] such as ecdsa-jcs-2019 and eddsa-jcs-2022 use the [[[?RFC8785]]] to [=transformation|transform=] the data to canonicalized JSON, which is then cryptographically hashed and digitally signed. One benefit of this approach is that adding or removing formatting characters that do not impact the meaning of the information being signed, such as spaces, tabs, and newlines, does not invalidate the digital signature. More traditional digital signature mechanisms do not have this capability.
Other [=cryptosuites=] such as ecdsa-rdfc-2019 and eddsa-rdfc-2022 use [[[?RDF-CANON]]] to [=transformation|transform=] the data to canonicalized N-Quads [[?N-QUADS]], which is then cryptographically hashed and digitally signed. One benefit of this approach is that the cryptographic signature is portable to a variety of different syntaxes, such as JSON, YAML, and CBOR, without invalidating the signature. More traditional cryptographic signature mechanisms do not have this capability.
Implementers and developers are urged to not trust information that contains a [=data integrity proof=] unless the proof has been [=verified=] and the verified data is provided in a return value from a software library that has confirmed that all data returned has been successfully protected.
The inspectability of application data has effects on system efficiency and developer productivity. When cryptographically protected application data, such as base-encoded binary data, is not easily processed by application subsystems, such as databases, it increases the effort of working with the cryptographically protected information. For example, a cryptographically protected payload that can be natively stored and indexed by a database will result in a simpler system that:
Similarly, a cryptographically protected payload that can be processed by multiple upstream networked systems increases the ability to properly layer security architectures. For example, if upstream systems do not have to repeatedly decode the incoming payload, it increases the ability for a system to distribute processing load by specializing upstream subsystems to actively combat attacks. While a digital signature needs to always be checked before taking substantive action, other upstream checks can be performed on transparent payloads — such as identifier-based rate limiting, signature expiration checking, or nonce/challenge checking — to reject obviously bad requests.
Additionally, if a developer is not able to easily view data in a system, the ability to easily audit or debug system correctness is hampered. For example, requiring application developers to cut-and-paste base-encoded application data makes development more challenging and increases the chances that obvious bugs will be missed because every message needs to go through a manually operated base-decoding tool.
There are times, however, where the correct design decision is to make data opaque. Data that does not need to be processed by other application subsystems, as well as data that does not need to be modified or accessed by an application developer, can be serialized into opaque formats. Examples include digital signature values, cryptographic key parameters, and other data fields that only need to be accessed by a cryptographic library and need not be modified by the application developer. There are also examples where data opacity is appropriate when the underlying subsystem does not expose the application developer to the underlying complexity of the opaque data, such as databases that perform encryption at rest. In these cases, the application developer continues to develop against transparent application data formats while the database manages the complexity of encrypting and decrypting the application data to and from long-term storage.
This specification strives to provide an architecture where application data remains in its native format and is not made opaque, while other cryptographic data, such as digital signatures, are kept in their opaque binary encoded form. Cryptographic suite implementers are urged to consider appropriate use of data opacity when designing their suites, and to weigh the design trade-offs when making application data opaque versus providing access to cryptographic data at the application layer.
Implementers must ensure that a verification method is bound to a particular controller by going from the verification method to the controller document, and then ensuring that the controller document also contains the verification method.
When an implementation is verifying a proof, it is imperative that it verify not only that the [=verification method=] used to generate the proof is listed in the [=controller document=], but also that it was intended to be used to generate the proof that is being verified. This process is known as "verification relationship validation".
The process of validating a verification relationship is outlined in Section 3.3 Retrieve Verification Method of the [[[CONTROLLER-DOCUMENT]]] specification.
This process is used to ensure that cryptographic material, such as a private cryptographic key, is not misused by application to an unintended purpose. An example of cryptographic material misuse would be if a private cryptographic key meant to be used to issue a Verifiable Credential was instead used to log into a website (that is, for authentication). Not checking a verification relationship is dangerous because the restriction and protection profile for some cryptographic material could be determined by its intended use. For example, some applications could be trusted to use cryptographic material for only one purpose, or some cryptographic material could be more protected, such as through storage in a hardware security module in a data center versus as an unencrypted file on a laptop.
When an implementation is verifying a proof, it is imperative that it verify that the [=proof purpose=] match the intended use.
This process is used to ensure that proofs are not misused by an application for an unintended purpose, as this is dangerous for the proof creator. An example of misuse would be if a proof that stated its purpose was for securing assertions in [=verifiable credentials=] was instead used for [=authentication=] to log into a website. In this case, the proof creator attached proofs to any number of [=verifiable credentials=] that they expected to be distributed to an unbounded number of other parties. Any one of these parties could log into a website as the proof creator if the website erroneously accepted such a proof as [=authentication=] instead of its intended purpose.
The way in which a transformation, such as canonicalization, is performed can affect the security characteristics of a system. Selecting the best canonicalization mechanisms depends on the use case. Often, the simplest mechanism that satisfies the desired security requirements is the best choice. This section attempts to provide simple guidance to help implementers choose between the two main canonicalization mechanisms referred to in this specification, namely JSON Canonicalization Scheme [[RFC8785]] and RDF Dataset Canonicalization [[RDF-CANON]].
If an application only uses JSON and does not depend on any form of RDF semantics, then using a cryptography suite that uses JSON Canonicalization Scheme [[RFC8785]] is an attractive approach.
If an application uses JSON-LD and needs to secure the semantics of the document, then using a cryptography suite that uses RDF Dataset Canonicalization [[RDF-CANON]] is an attractive approach.
Implementers are also advised that other mechanisms that perform no transformations are available, that secure the data by wrapping it in a cryptographic envelope instead of embedding the proof in the data, such as JWTs [[?RFC7519]] and CWTs [[?RFC8392]]. These approaches have simplicity advantages in some use cases, at the expense of some of the benefits provided by the approach detailed in this specification.
One of the algorithmic processes used by this specification is canonicalization, which is a type of [=transformation=]. Canonicalization is the process of taking information that might be expressed in a variety of semantically equivalent ways as input, and expressing all output in a single way, called a "canonical form".
The security of a resulting [=data integrity proof=] that utilizes canonicalization is highly dependent on the correctness of the algorithm. For example, if a canonicalization algorithm converts two inputs that have different meanings into the same output, then the author's intentions can be misrepresented to a [=verifier=]. This can be used as an attack vector by adversaries.
Additionally, if semantically relevant information in an input is not present in the output, then an attacker could insert such information into a message without causing proof verification to fail. This is similar to another transformation that is commonly used when cryptographically signing messages: cryptographic hashing. If an attacker is able to produce the same cryptographic hash from a different input, then the cryptographic hash algorithm is not considered secure.
Implementers are strongly urged to ensure proper vetting of any canonicalization algorithms to be used for [=transformation=] of input to a [=hashing=] process. Proper vetting includes, at a minimum, association with a peer reviewed mathematical proof of algorithm correctness; multiple implementations and vetting by experts in a standards setting organization is preferred. Implementers are strongly urged not to invent or use new mechanisms unless they have formal training in information canonicalization and/or access to experts in the field who are capable of producing a peer reviewed mathematical proof of algorithm correctness.
This specification is designed in such a way that no network requests are required when verifying a proof on a [=conforming secured document=]. Readers might note, however, that JSON-LD contexts and [=verification methods=] can contain URLs that might be retrieved over a network connection. This concern exists for any URL that might be loaded from the network during or after verification.
To the extent possible, implementers are urged to permanently or aggressively cache such information to reduce the attack surface on an implementation that might need to fetch such URLs over the network. For example, caching techniques for JSON-LD contexts are described in Section [[[#contexts-and-vocabularies]]], and some [=verification methods=], such as `did:key` [[?DID-KEY]], do not need to be fetched from the network at all.
When it is not possible to use cached information, such as when a specific HTTP URL-based instance of a [=verification method=] is encountered for the first time, implementers are cautioned to use defensive measures to mitigate denial-of-service attacks during any process that might fetch a resource from the network.
Since the technology to secure documents described by this specification is generalized in nature, the security implications of its use might not be immediately apparent to readers. To understand the sort of security concerns one might need to consider in a complete software system, implementers are urged to read about how this technology is used in the [=verifiable credentials=] ecosystem [[?VC-DATA-MODEL-2.0]]; see the section on Verifiable Credential Security Considerations for more information.
The following section describes privacy considerations that developers implementing this specification should be aware of in order to create privacy enhancing software.
When a digitally-signed payload contains data that is seen by multiple verifiers, it becomes a point of correlation. An example of such data is a shopping loyalty card number. Correlatable data can be used for tracking purposes by verifiers, which can sometimes violate privacy expectations. The fact that some data can be used for tracking might not be immediately apparent. Examples of such correlatable data include, but are not limited to, a static digital signature or a cryptographic hash of an image.
It is possible to create a digitally-signed payload that does not have any correlatable tracking data while also providing some level of assurance that the payload is trustworthy for a given interaction. This characteristic is called unlinkability which ensures that no correlatable data are used in a digitally-signed payload while still providing some level of trust, the sufficiency of which must be determined by each verifier.
It is important to understand that not all use cases require or even permit unlinkability. There are use cases where linkability and correlation are required due to regulatory or safety reasons, such as correlating organizations and individuals that are shipping and storing hazardous materials. Unlinkability is useful when there is an expectation of privacy for a particular interaction.
There are at least two mechanisms that can provide some level of unlinkability. The first method is to ensure that no data value used in the message is ever repeated in a future message. The second is to ensure that any repeated data value provides adequate herd privacy such that it becomes practically impossible to correlate the entity that expects some level of privacy in the interaction.
A variety of methods can be used to achieve unlinkability. These methods include ensuring that a message is a single use bearer token with no information that can be used for the purposes of correlation, using attributes that ensure an adequate level of herd privacy, and the use of cryptosuites that enable the entity presenting a message to regenerate new signatures while not compromising the trust in the message being presented.
Selective disclosure is a technique that enables the recipient of a previously-signed message (that is, a message signed by its creator) to reveal only parts of the message without disturbing the verifiability of those parts. For example, one might selectively disclose a digital driver's license for the purpose of renting a car. This could involve revealing only the issuing authority, license number, birthday, and authorized motor vehicle class from the license. Note that in this case, the license number is correlatable information, but some amount of privacy is preserved because the driver's full name and address are not shared.
Not all software or cryptosuites are capable of providing selective disclosure. If the author of a message wishes it to be selectively disclosable by its recipient, then they need to enable selective disclosure on the specific message, and both need to use a capable cryptosuite. The author might also make it mandatory to disclose certain parts of the message. A recipient that wants to selectively disclose partial content of the message needs to utilize software that is able to perform the technique. An example of a cryptosuite that supports selective disclosure is `bbs-2023`.
It is possible to selectively disclose information in a way that does not preserve unlinkability. For example, one might want to disclose the inspection results related to a shipment, which include the shipment identifier or lot number, which might have to be correlatable due to regulatory requirements. However, disclosure of the entire inspection result might not be required as selectively disclosing just the pass/fail status could be deemed adequate. For more information on disclosing information while preserving privacy, see Section [[[#unlinkability]]].
When using the `previousProof` feature defined in [[[#proof-chains]]], implementations are required to digitally sign over one or more previous proofs, so as to include them in the secured payload. This inevitably exposes information related to each entity that added a previous proof.
At minimum, the [=verification method=] for the previous proof, such as a public key, is seen by the creator of the next proof in a proof chain. This can be a privacy concern if the creator of the previous proof did not intend to be included in a proof chain, but is an inevitable outcome when adding a non-repudiable digital signature to a document of any kind.
It is possible to use more advanced cryptographic mechanisms, such as a group signature, to hide the identity of the signer of a message, and it is also possible for a Data Integrity cryptographic suite to mitigate this privacy concern.
Fingerprinting concerns exist for any URL that might be loaded from the network during or after proof verification. This specification is designed in such a way that no network requests are necessary when verifying a proof on a [=conforming secured document=]. Readers might note, however, that JSON-LD contexts and [=verification methods=] can contain resource URLs that might be retrieved over a network connection leading to fingerprinting concerns.
For example, creators of [=conforming secured documents=] might craft unique per-document URLs for JSON-LD contexts and [=verification methods=]. When verifying such a document, a verifier fetching that information from the network would reveal their interest in the [=conforming secured document=] to the creator of the document, which might lead to a mismatch in privacy expectations for any entity that is not the creator of the document.
Implementers are urged to follow the guidance in Section [[[#network-requests]]] on URL caching and implementing defensively when fetching URLs from the network. Usage of techniques such as Oblivious HTTP to retrieve resources from the network, without revealing the client that is making the request, are encouraged. Additionally, heuristics might be used to determine whether creators of [=conforming secured documents=] are using fingerprinting URLs in a way that might violate privacy expectations. These heuristics could be used to display warnings to entities that might process documents containing suspected fingerprinting URLs.
The way in which a transformation, namely canonicalization, is performed can affect the privacy characteristics of a system. Selecting the best canonicalization mechanism depends on the use case. This section attempts to provide simple guidance to help implementers pick between the two main canonicalization mechanisms referred to in this specification, namely JSON Canonicalization Scheme [[RFC8785]] and RDF Dataset Canonicalization [[RDF-CANON]], from a privacy perspective.
If an application does not require performing a selective disclosure of information in a secured document, nor does it utilize JSON-LD, then JSON Canonicalization Scheme [[RFC8785]] is an attractive approach.
If an application uses JSON-LD and might require selective disclosure of information in a secured document, then using a cryptography suite that uses RDF Dataset Canonicalization [[RDF-CANON]] is an attractive approach.
Implementers are also advised that other selective disclosure mechanisms that perform no transformations are available, that secure the data by wrapping it in a cryptographic envelope instead of embedding the proof in the data, such as SD-JWTs [[?SD-JWT]]. This approach has simplicity advantages in some use cases, at the expense of some of the benefits provided by the approach detailed in this specification.
Since the technology to secure documents described by this specification is generalized in nature, the privacy implications of its use might not be immediately apparent to readers. To understand the sort of privacy concerns one might need to consider in a complete software system, implementers are urged to read about how this technology is used in the [=verifiable credentials=] ecosystem [[?VC-DATA-MODEL-2.0]]; see the section on Verifiable Credential Privacy Considerations for more information.
The following section describes accessibility considerations that developers implementing this specification are urged to consider in order to ensure that their software is usable by people with different cognitive, motor, and visual needs. As a general rule, this specification is used by system software and does not directly expose individuals to information subject to accessibility considerations. However, there are instances where individuals might be indirectly exposed to information expressed by this specification and thus the guidance below is provided for those situations.
This specification enables the expression of dates and times related to the validity period of cryptographic proofs. This information might be indirectly exposed to an individual if a proof is processed and is detected to be outside an allowable time range. When exposing these dates and times to an individual, implementers are urged to take into account cultural normas and locales when representing dates and times in display software. In addition to these considerations, presenting time values in a way that eases the cognitive burden on the individual receiving the information is a suggested best practice.
For example, when conveying the expiration date for a particular set of digitally signed information, implementers are urged to present the time of expiration using language that is easier to understand rather than language that optimizes for accuracy. Presenting the expiration time as "This ticket expired three days ago." is preferred over a phrase such as "This ticket expired on July 25th 2023 at 3:43 PM." The former provides a relative time that is easier to comprehend than the latter time, which requires the individual to do the calculation in their head and presumes that they are capable of doing such a calculation.
Sections [[[#proof-sets]]] and [[[#proof-chains]]] describe how multiple proofs can be expressed in a [=secured data document=]; that is, instead of a single [=proof=] included in the [=secured data document=], one can express multiple proofs in an array as shown in [[[#example-a-proof-set-in-a-data-document]]] and [[[#example-a-proof-chain-in-a-data-document]]]. The elements of this array are members of a [=proof set=] and, optionally, a [=proof chain=]. The purpose of this section is to explain the intended use of each of these features and, in particular, their differing security properties. These differing security properties lead to differences in the processing in section [[[#add-proof-set-chain]]].
This section represents [=secured data documents=], including their proofs, in an abbreviated manner so that the important security properties can be observed.
Consider a scenario with three signatories: a CEO, a CFO, and a VP of Engineering. Each will need to have a public key and secret key pair for signing a document. We denote the secret/public keys of each of these signatories by secretCEO/publicCEO, secretCFO/publicCFO, and secretVPE/publicVPE, respectively.
When constructing a [=proof set=] where each of the signatories signs an |unsecuredDataDocument| without concern, we construct a proof symbolically as:
{ "type": "DataIntegrityProof", "cryptosuite": "eddsa-jcs-2022", "created": "2023-03-05T19:23:24Z", "proofPurpose": "assertionMethod", "verificationMethod":publicCEO
, "proofValue": signature(secretCEO
,unsecuredDataDocument
) }
Where publicCEO is used as a placeholder for a reference that resolves to the CEO's public key and signature(`secretKey`, `unsecuredDataDocument`) denotes the computation of a digital signature by a particular data integrity cryptosuite using a particular secret key over a particular document. The `type`, `cryptosuite`, `created`, and `proofPurpose` attributes do not factor into our discussion so we will omit them. In particular, below we show all the proofs in a [=proof set=] on a document that has been signed by the VP of Engineering, the CFO, and the CEO:
{ // Remainder of secured data document not shown (above) "proof": [{ "verificationMethod":publicVPE
, "proofValue": signature(secretVPE
,unsecuredDataDocument
) }, { "verificationMethod":publicCFO
, "proofValue": signature(secretCFO
,unsecuredDataDocument
) }, { "verificationMethod":publicCEO
, "proofValue": signature(secretCEO
,unsecuredDataDocument
) }] }
A [=holder=] or any other intermediary receiving a [=secured data document=] containing a [=proof set=] is able to remove any of the `proof` values within the set prior to passing it on to another entity and the [=secured data document=] will still verify. This might or might not have been the intent. For the signatories sending a birthday card to a valued employee, using a [=proof set=] is probably fine. If we are trying to model a business process where approvals ascend the company hierarchy, this would not be ideal, since any intermediary could remove signatures from the [=proof set=] and still have it verify; for instance, in the example below, it looks like the CFO and CEO approved something without the VP of Engineering's concurrence.
{ // Remainder of secured data document not shown (above) "proof": [{ "verificationMethod":publicCFO
, "proofValue": signature(secretCFO
,unsecuredDataDocument
) }, { "verificationMethod":publicCEO
, "proofValue": signature(secretCEO
,unsecuredDataDocument
) }] }
It is possible to introduce a dependency between [=proofs=] in a [=proof set=] by setting the `id` property of each proof such that another proof can reference it. In other words, a dependent proof will be referenced by other relying proofs by using the `previousProof` property. Such dependency chains can have arbitrary depth. The intent of such a [=proof chain=] is to model an approval chain in a business process or a notary witnessing analog signatures.
The examples below demonstrate how a [=proof chain=] can be constructed when the VP of Engineering signs off on the document first; based on the VP of Engineering's signature and a review, the CFO then signs off on the document; and finally, based on both prior signatures and a review, the CEO signs off on the document. Since others will be referring to the VP of Engineering's signature, we need to add an `id` to the proof. First the VP of Engineering signs the [=unsecured data document=]:
{ // Remainder of secured data document not shown (above) "proof": { "id": "urn:proof-1", "verificationMethod":publicVPE
, "proofValue": signature(secretVPE
,unsecuredDataDocument
) } }
Next, the CFO receives the document, verifies that the VP of Engineering signed it, and signs it based on a review and on the signature of the VP of Engineering. For this, we need to set up the [=proof chain=] by indicating a dependency on the proof in the document just received. We do this by setting the `previousProof` property of the second proof to the value `urn:proof-1`, which "binds" the second proof to the first proof, which is then signed. The following example shows how the dependency on the first proof is created:
{ // Remainder of secured data document not shown (above) "proof": [{ "id": "urn:proof-1", "verificationMethod":publicVPE
, "proofValue": signature(secretVPE
,unsecuredDataDocument
) }, { "id": "urn:proof-2", "verificationMethod":publicCFO
, "previousProof": "urn:proof-1", "proofValue": signature(secretCFO
,unsecuredDataDocumentWithProof1
) }] }
Now, when the CEO verifies the received [=secured data document=] with the above [=proof chain=], they will check that the CFO signed based on the signature of the VP of Engineering. First, they will check the proof with an `id` property whose value is `urn:proof-1` against the public key of the VP of Engineering. Note that this proof is over the original document.
Next, the CEO will check the proof with an `id` property whose value is `urn:proof-2` against the public key of the CFO. However, to make sure that the CFO signed the document with proof that the VP of Engineering had already signed, we verify this proof over the combination of the document and `urn:proof-1`. If verification is successful, the CEO signs, producing a proof over the document which includes `urn:proof-1` and `urn:proof-2`. The final [=proof chain=] looks like this:
{ // Remainder of secured data document not shown (above) "proof": [{ "id": "urn:proof-1", "verificationMethod":publicVPE
, "proofValue": signature(secretVPE
,unsecuredDataDocument
) }, { "id": "urn:proof-2", "verificationMethod":publicCFO
, "previousProof": "urn:proof-1", "proofValue": signature(secretCFO
,unsecuredDataDocumentWithProof1
) }, { "id": "urn:proof-3", "verificationMethod":publicCEO
, "previousProof": "urn:proof-2", "proofValue": signature(secretCEO
,unsecuredDataDocumentWithProof2
) }] }
The recipient of this [=secured data document=] then validates it in a similar way, checking each proof in the chain.
This section contains the substantive changes that have been made to this specification over time.
Changes since the First Public Working Draft: