Skip to content

Latest commit

 

History

History
760 lines (656 loc) · 44.1 KB

selective-correlation.md

File metadata and controls

760 lines (656 loc) · 44.1 KB
Raw

Selective Correlation

Data Minimization, Selective Disclosure, and Progressive Trust

Authors

Introduction

The goal of this paper is to enable decision makers, particularly non-technical ones, to gain a nuanced grasp of correlation. We will explore the concept of selective correlation, introduce a threat model, and propose a framework within which policies such as Data Minimization can be successfully actualized. We will strive to do so in plain English, but with some rigor. This will enable readers of this paper to better understand correlation best practices, to adopt correct approaches for their use cases, and to assess techniques that enable these best practices.

Key Concepts

Correlation and Identity

Correlation lies at the heart of identity. The concept of identity cannot exist without some observer who identifies by correlating different data points about the subject being identified, whether that subject is a person, group, or thing. Identification is correlation across contexts: the subject being identified is recognized as the same subject from a different context. Control over correlation is therefore directly tied to control of identity, or in other words, tied to control of how identifiable the subject is. 1

Correlation is a required function of any identity system. However, correlation historically also has been misused by adversaries to threaten or harm people. To prevent the misuse of correlation in the context of self-sovereign identity2, the power to correlate must also be self-sovereign. For individuals this means that the power to correlate must be, as much as possible, in the hands of the person being correlated.

In most systems, people have little control about what information they provide. Much of the data that allows them to be correlated lies outside their control. In a self-sovereign system, people have more opportunities to manage correlatable data.

Correlatable Data

We define correlatable data as any data than can be used to identify a subject across contexts. A context may be siloed, as when data is shared with two different observers. Or it may be temporal, as when data is shared with the same observer over time.

We define an observer as an entity who collects correlatable data about a subject. An observer may collect data directly from subjects, from the process used to collect data about subjects, or from other observers. In the context of digital credentials, an observer may aso be called a verifier.

The more correlatable data that is available, the easier it will be for an observer to identify subjects without their direct consent. Therefore, a reduction in the quantity of available correlatable data is desirable. One means of accomplishing this is to follow the practice of data minimization.

Data Minimization

Data minimization is the practice of limiting the shared data to the minimum amount necessary. What this means concretely depends upon context, but in general it means only sharing enough data for parties to successfully transact, while minimizing all unnecessary data.

Though essential as part of security best practices, minimal disclosure of data is vital for maintaining the privacy of people using digital identities. Special attention must be paid when collecting and sharing data, to limit that sharing to only that which is absolutely necessary.

The concept of data minimization guides the design and implementation of personal data protection regulations, such as the General Data Protection Regulation (GDPR) in the European Union. For example:

  • Service providers must only collect the minimum amount of personal information necessary to perform a specific task or service.
  • Service providers must imit the length of time personal information is retained and must delete it when it is no longer needed.
  • Individuals should only share personal information with third parties when it is necessary to perform a task or service.

Other regulations that require some form of data minimization are listed in references.

ISO (International Organization for Standardization) has several standards related to data protection and privacy that include principles of data minimization. We list some in references.

In general, data minimization is enacted by policy decisions that limit the amount, duration, and scope of data collection and use:

  • Content minimization (amount) involves collecting only the minimum amount of data necessary.
  • Temporal minimization (duration) involves retaining data only for the minimum amount of time necessary to execute the task.
  • Scope minimization (scope) involves using data only for the specific purpose of the active task.

For digital credentials, data minimization is primarily implemented through policy decisions made by stakeholders: credential issuers ensure that credentials can be presented in a way that enables data minimization, and credential verifiers establish policies in advance regarding the minimum data necessary to accomplish a task, the minimum time data can be stored, and the processes that ensure data is only applied to the task at hand.

In addition to the use of data minimization to protect the privacy of data subjects, verifiers also benefit from data minimization. These benefits include avoiding collection of "toxic" personal data and reducing the cost of storing information.

The implementation of data minimization can be challenging:

  • Determining the minimum necessary data: It can be difficult to determine exactly how much data is necessary, especially when dealing with complex systems and processes.
  • Balancing data minimization with other goals: Organizations may be torn between the need to collect and retain data for legitimate business purposes and the need to minimize data to protect privacy and security.
  • Siloed data: Data may be collected and stored across multiple systems and departments, making it difficult to identify and remove unnecessary data.
  • User consent: People are not willing to share their personal information.

Terminology

Correlatable: Data is said to be correlatable if by examining it that there is a way to determine whether it is associated with sets of other data stored elsewhere. This also includes lossy projections of the sets of data such as deterministic hashes that demonstrate that assocation.

Decorrelation: A general term for any process that is used to reduce correlation within data, or cross-correlation with other sets of data, while preserving other useful aspects of the data.

Differential Privacy: A decorrelation process for sharing information about sets of data by describing the patterns of groups within the dataset while withholding information about individuals in the dataset. The idea behind differential privacy is that if the effect of making an arbitrary single substitution in the database is small enough, the query result cannot be used to infer much about any single individual, and therefore provides privacy. For instance, adding a random number (say 4) from one part of the set and subtracting 4 from another part of the set when the business purpose of the data is total or average.

Elide/Elision: The term elide means "to leave out", and elision is the act or instance of ommitting something. (Redaction is a related term

Herd Privacy: The decorrelation choices made by one set vs other data sets may result in correlation. Herd privacy process ensures that this doesn't happen by making them indistiguishable from other data sets.

Non-correlatable: If there is no practical way to learn whether a set of data is a projection of a other data, they are said to be noncorrelatable.

Nonce: aka "Number Used Once". This is an arbitrary number that is used just once in a cryptographic function such as signatures or encryption, to prevent correlation. Note that this is not necessarly a random number, simply a number that is never used again, and in some cases it can be quite valuable for a nonce to be generated deterministically.

Quasi-correlatable: Between projections that are definitely correlatable and definitely noncorrelatable, there are projections that may leak a little information about their data. In particular (in no particular order), size, order, use of particular formatting, date/time, location, algorithm usage, and other identifiable patterns.

Salt: This is random data (values and length) that are used as an additional input to a cryptographic function such as a hash of data, encryption, signature, to prevent correlation.

Techniques

Elision

Elision is the ability of an individual to derive from a credential another credential with some attributes removed. Stakeholders in the credentials ecosystem enable elision capabilities in the following ways:

  • Credential issuers negotiate with the holder to enable elision, by issuing credentials with a subset of attributes already included in another credential.
  • Credential inspectors ensure that the elided credentials can be verified with the usual procedure, but they won't be able to know if any kind of elision has been performed if they cannot correlate them with the starting credentials.

Selective disclosure

Selective disclosure is the ability of an individual to granularly decide what information to share. Stakeholders in the credentials ecosystem enable selective disclosure capabilities in the following ways:

  • Credential issuers format the credential and its attributes in such a way as to enable selective disclosure. As with the strategy of data minimization, they may issue multiple, related, granular sub-credentials. Each attribute and the overall credential may be formatted to support cryptography, a capability described in more detail below.
  • Credential inspectors ensure the request is framed in such a way as to enable selective disclosure, using the cryptographic tools required.

Once data minimization policies and selective disclosure are in place, the third and last enhancement can be applied.

Progressive Trust

Progressive trust is the process usable by an individual to gradually increase the amount of relevant data revealed as trust is built or value generated.

To enable progressive trust capabilities, stakeholders in the credentials ecosystem act in the following ways:

  • Issuers format the credential(s) in such a way as to enable progressive trust. This may require issuing multiple, related, atomic sub-credentials. It also may require formatting the credential to support mathematical queries and cryptographic proofs. Finally, the issuer's data model may express how the various sub-credentials are related in a scenario involving progressive trust.
  • Inspectors ensure that requests are framed in such a way as to enable progressive trust. They structure the communication in order to to gradually escalate credential requests in order to enable a subject to progressively trust the inspector more and more, revealing the minimum data necessary to accomplish each step of the task or goal and revealing more and more as the mutual communication progresses.

Progressive Disclosure

Progressive disclusure is one of the possible ways for an individual to let the progressive trust with another party proceed. The individual starts anonymously and continues to provide information, recognizing that the more data is revealed, the more risk that the other would be able to correlate. But with the gained trust, the risk will be considered a benefit, instead.

To enable progressive disclosure capabilities, stakeholders in the credentials ecosystem act in the following ways:

  • Issuers format the credential(s) in such a way as to enable progressive disclosure. This requires issuing credentials that can be selectively disclosed multiple times. Each presentation will share attributes that have not been shared before.
  • Inspectors ensure that requests are framed in such a way as to enable progressive trust. This requires presentation protocols that can link together some previously made presentations, forming a progressive trust instance.

Selective Disclosure

"Selective Disclosure" is one of a number of privacy-enhancing techniques that go beyond Data Minimization to also protect against correlation. Selective Disclosure allows individuals or organizations to share only specific pieces of information, rather than sharing all of them, and prevents using correlation inappropriately to merge information from different contexts without consent. Selective Disclosure offers approaches that balance the need to share information for legitimate purposes against the need to protect the privacy and security of people and to minimize risks.

GDPR does not mandate the use of Selective Disclosure, but the GDPR does gives individuals the right to control their personal information, the right to access, correct, and delete their personal information, and the right to object to certain processing activities. The California Consumer Privacy Act (CCPA) also allows individuals to know what personal information is being collected and shared and to opt out of the sale of their personal information.

In addition, some organizations have developed their own standards for Selective Disclosure, such as the Platform for Privacy Preferences (P3P) which provides a mechanism for websites to disclose their data collection and sharing practices and for individuals to set their privacy preferences.

Some requirements for Selective Disclosure include:

  • Granularity: Allow individual or organizations to share only specific pieces of information, rather than sharing all of their personal information. This enables users to share only the information that is necessary to accomplish a specific task or goal.
  • Control: Give users more control over their personal information by allowing them to decide what information they want to share, and with whom they want to share it.
  • Transparency: Allow users to see what information is being shared and with whom, which enhances trust and transparency in the sharing process.
  • Security: Use cryptographic techniques to secure the information shared, ensuring that only authorized individuals or organizations can access the information.
  • Privacy: Minimize the amount of personal information that is shared, reducing the risk of data breaches or unauthorized access to personal information.
  • Compliance: Help organizations to comply with data protection regulations, such as GDPR, by minimizing the amount of personal data collected, used and retained.
  • Auditability: Allow organizations to track and audit information-sharing activities, providing transparency on how the data is being used.
  • Flexibility: Allow organizations to adapt the sharing process to different scenarios and use cases, providing necessary flexibility.

More specifically, Selective Disclosure can help address some of the challenges of Data Minimization mentioned above:

  • Difficulty in determining the minimum necessary data: By allowing users to share only the specific information needed to accomplish a task or goal, Selective Disclosure can limit the data collection to only necessary information, avoiding additional information that is sensitive and potentially toxic.
  • Balance of data minimization with other goals: By requiring granulatity in Selective Disclosure, the different goals of different groups can be evaluated and addressed.
  • Usage of data silos: Selective Disclosure can help with data silos by allowing users to share information from different systems and departments while ensuring that only authorized individuals or organizations can access the information and that it is not shared or stored for longer than necessary. This can also help organizations to comply with data protection regulations such as GDPR.
  • Lack of user consent: By giving users more control over their personal information, Selective Disclosure allows organizations to obtain user consent for information sharing in a more effective way. Users can choose what information they want to share and with whom, giving them control over what information an organization has access to. This can also increase trust and transparency between the organization and its users.

Cryptographic Techniques for Selective Disclosure

Cryptographic selective disclosure leverages cryptography to allow individuals to selectively share specific pieces of their information while keeping the rest of their information private. It allows parties to prove certain attributes about themselves, without revealing their entire identity.

There are three important approaches to cryptographic selective disclosure:

  • "Hash-based Elision (or Redaction)": A hash is cryptographic fingerprint of a set of data — it takes an input and creates a unique output. In Hash-based Elision, one party (the prover) presents to another party (the verifier) a hash of a piece of personal information (redacts it), without revealing the actual information. One advantage of using cryptographic hashes for selective disclosure is that they are relatively simple and efficient method for hiding personal information, don't require complex mathematical calculations, and it doesn't require a trusted third party. A disadvantage of using cryptographic hashes is that without multiple round-trips they do not provide any cryptographic guarantees that the prover knows the original data (they may be just passing the hash forward). Another disadvantage is that there is a correlation risk if the same hash is given to multiple parties, which requires techiques like salting.

  • "Zero-Knowledge proof (ZKP): This cryptographic technique allows one party (the prover) to prove to another party (the verifier) that they possess certain information, without revealing the actual information. This is done by allowing the prover to demonstrate that they know the information, without revealing the actual data. ZKP can be used to prove a wide range of attributes like knowledge of a password, possession of a private key, and membership of a group, etc. As compared to cryptographic hashes, ZKP is a method for proving knowledge of a subset of information without revealing the actual information behind it, while cryptographic hashes are a way to hide part of the actual information, but not the knowledge of it. A disadvantage of ZKP is that it can be computationally expensive, or requires a large amount of communication between the prover and verifier, both can be a bottleneck in certain scenarios where low latency is a concern or on constrained hardware.

  • Blind Signature: As the name implies, this is a technique in which a signature is "blinded" before it is signed, hiding the identity of the signer. This allows the signer to prove that they possess a certain attribute, such as being over 18 years old, without revealing their name or other personal information. The signature can then be "unblinded" to reveal the identity of the signer. This makes it useful in scenarios where it is important to prove that a document has been signed by a specific individual without revealing their identity. One disadvantage of blind signatures is that they rely on a trusted third party, which can be a bottleneck and a centralization or security risk.

There are other possible approaches to leverage cryptography for selective disclosure, but these are not as broadly being investigated today.

  • Secret Sharing: This technique allows a secret to be divided into specific number of shares, sent to to parties, from which a specific number of shares are required (a quorum) to reconstruct the secret. A disadvantage is that restoring original secret requires one party to be entrusted to restore the quorum.
  • Secure Multi-Party Computation (MPC): This technique allows multiple parties to jointly compute a function without revealing their inputs to each other. This solves the problems inherent to secret sharing, but at the cost of multiple rounds of interaction between the parties, and computational complexity.
  • Homomorphic encryption: This technique allows computations to be performed on ciphertext, resulting in the same plaintext output as if the computation was performed on the plaintext. This y allowing computations to be performed on encrypted data without decrypting it first or revealing the actual data. However, these techniques are extremely computationally expensive (by multiple orders of magnitude).

Use Cases

These use cases focus at their core a single credential, presented by various parties to others, with desirable correlation as well as threats if undesirable correlation occurs. An actual decentralized system probably has more trust assurance, multiple credentials and more verifiable presentation metadata than what is discussed below.

Educational / Professional Use Case:

The Educational Use Cases are drawn in part from https://github.com/BlockchainCommons/Gordian/blob/master/Docs/Envelope-Use-Cases-Educational.md, which was itself informed by discussions related to this paper.

  1. DANIKA ENTERS SCHOOL: Danika, a citizen of the Czech Republic, desires comprehensive training in a highly paid building construction specialty: plasma welding. She applies to Czech-PAW, a professional trade school, and is accepted. Danika offers Czech-PAW a DID, and adds details such as a photo that will allow others to authenticate her using a Student ID. Students of Czech-PAW are also qualified to receive students loans from Raiffeisenbank, so Danika applies and receives a loan, and payments are deferred until she graduates.

    • TRANSACTIONS: Danika becomes a student of Czech-PAW. Danika receives Student ID. Danika applies for student loan and pays for schooling.
    • DESIRABLE CORRELATION: Danika desires to be correlated as a student to attend classes and events, to get a loan to pay for classes, and to access other benefits such as student discounts.
    • CORRELATION THREAT: Danika might wish to keep her studies private from a current employer.

  2. DANIKA GETS AN EDUCATIONAL CREDENTIAL: Danika (the subject) obtains comprehensive training from Czech-PAW (the issuer) on plasma welding, including a number of advanced techniques along with a variety of workplace safety topics. Czech-PAW issues an education credential to Danika (the subject & holder) with all of these details that confirm Danika has all the proper training and experience for this complex technical discipline.

    • TRANSACTIONS: Danika completes schooling. Danika finishes payment for schooling. Danika receives educational credential from Czech-PAW.
    • DESIRABLE CORRELATION: Danika desires to be correlated as a graduate of school and qualified for various job prospects.
    • CORRELATION THREAT: Danika may wish to keep her educational credentials at Czech-PAW separate from other educational credentials, lest the attainment of "blue collar" professional credentials be a barrier to other sorts of jobs.

  3. EXAMPLE OF ELISION BY A DANIKA (SUBJECT/HOLDER): Danika moves to the United States, when she applies for a plasma welding gig at Erickson's Construction in New York. They require a workplace safety certification. Danika (the subject & holder) offers her Czech-PAW educational credential with details on the training and safety certification (the desired correlation as being qualified for the job) but elides her name and the name of institute, as she's learned from prior applications that revealing her Eastern European origins can cause discrimination. She can still prove ownership of the certification by signing with her DID, and she can still prove the accreditation of her trade school with EU-level certifications.

    • TRANSACTIONS: Danika applies for employment with educational credential.
    • DESIRABLE CORRELATION: Danika desires to be correlated as a qualified graduate and for her school to be correlated as a legitmate accredited school.
    • CORRELATION THREAT: Danika does not wish to be discriminated against for his familial name or for the location of her school.

  4. EXAMPLE OF PROGRESSIVE TRUST: Erickson's Construction offers Danika a temporary gig at a reasonable hourly rate, pending additional verification of identity at the workplace. Danika (subject & holder) presents her Czech-PAW (issuer) educational credential again, but this time with additional information (most notably, her last name!) so that Erickson's Construction (verifier) can issue a pass to enter the workplace and fill out a W2 for tax purposes. Danika must also provide certain United States tax credentials, including her Individual Taxpayer Identification Number (ITIN), to finalize that process.

    • TRANSACTIONS: Danika receives offer, accepts offer. Gives ITIN, legal name.
    • DESIRABLE CORRELATION: Danika presents her educational credential again, with less elision, to satisfy workplace requirements.
    • CORRELATION THREAT: Danika's ITIN could now be correlated with her Czech origins if Erickson's were to misuse the information; Danika must trust they will not. Nonethless, this is a real threat that has emerged due to the need to use government identification that does not support elision.

  5. EXAMPLE OF ELISION BY A HOLDER: Erickson's Construction (holder≠subject) needs to prove to Three Towers Corporation (verifier) that they have sufficient number of qualified employees for a contract. They provide Danika's (subject) educational credential (already partially elided by the subject) to Three Towers but they further elide all of Danika's personal information, including her full name and picture, to obey regulations regarding the distribution of PII and more pragmatically to prevent poaching by other firms. Three Towers can verify that the school (Czech-PAW but elided by Danika) is accredited thanks to the EU-level certifications, but they don't learn anything about her individually. The same is true for other elided credentials that Erickson's sens to Three Towers.

    • TRANSACTIONS: Erickson's Constructions (holder) presents Three Towers (verifier) multiple elided employee educational credentials.
    • DESIRABLE CORRELATION: Erickson's Construction is able to repackage and further elide the credentials to provide correlation that they have sufficient and properly trained staffing, from accredited schools.
    • CORRELATION THREAT: Erickson simply wishes to prove the count of its accredited staff, not to reveal their PII to a third party, which could be correlated.

  6. EXAMPLE OF HASH-BASED CORRELATION: Czech-PAW desires to renew their relationship with Raiffeisenbank to allow their students to continue to receive loans. To do so, they must provide information on their past and present students, such as Danika, so that Raiffeisenbank can assess that a sufficient number of their loan holders are graduating. However, under Europe's GDPR rules, Raiffeisenbank doesn't want to receive PII of non-loanholders, because it would be toxic. To resolve this problem, Czech-Paw produces a list of all of its graduates, all of its drop-outs over the last 10 years, and all of its students, but entirely elides each student's information, leaving behind only a hash derived from the student's PII. Raiffeisenbank can then create hashes from the PII that it already possesses for loan holders and compare that to the hashes in the elided lists. This will allow them to see which category each of their loan holders falls into, without either them receiving PII about non-loanholders or Czech-PAW receiving PII about non-students.

    • TRANSACTIONS: Czech-PAW (issuer) presents Raiffeisenbank (verifier) multiple elided student educational credentials. Danika's (subject) educational credential is one of those credentials, on the list of graduates.
    • DESIRABLE CORRELATION: Both Czech-PAW and Raiffeisenbank need to correlate educational credentials with loan holders.
    • CORRELATION THREAT: Raiffeisenbank wants to avoid accepting correlatable personal data of non-loanholders, while Czech-PAW wants to avoid accepting correlatable personal data of non-students.

  7. EXAMPLE OF HERD PRIVACY USING HASH-BASED CORRELATION: Czech-PAW allows for remote learning for continuing education, focusing on safety instructions and updates to codes, regulations, and laws, none of which require hands-on training like its core offerings do. However, the lack of in-person interaction can create a somewhat less secure environment for the transmission of the PII used in credentials and the quantity of students involved can create administrative problems in the creation and maintenance of individual credentials. Czech-PAW has resolved both of these issues with herd-privacy credentials. When Danika logs in to the continuing education, she supplies a new DID. When she and other students complete the course, Czech-PAW then creates a massive credential that contains the continuing education information for all of the supplied DIDs. However, each DID is elided and replaced with a hash: students are given individual "paths" to their personal DID, which they can combine with that DID to prove that a specific hash belongs to them, demonstrating their continuing education. Danika ultimately chooses not to: immediately following the course, Czech-PAW is involved in a massive scandal involving gas ionization, and she fears it might damage her career prospects if she was known to be still involved with the institute. Fortunately, the structure of the herd-privacy certificate ensures that no one can know that Danika is part of the certificate unless she reveals it herself (not even Czech-PAW, if Danika used the best practice of providing a new DID, since her continuing education didn't need to independently correlate to her original training).

    • TRANSACTIONS: Danika takes a continuing education course. Czech-PAW (issuer) produces a massive but elided credential following the course and provides Danika with enough information to prove her presence within. Danika (subject) chooses whether to reveal that she is part of the certificate or not.
    • DESIRABLE CORRELATION: If Danika opts to prove her presence in the mass credential, it will correlate her continuing education to her existing credentials.
    • CORRELATION THREAT: However, the mass credential can also correlate Danika's continuing interactions with Czech-PAW, which she ultimately decides to avoid.

Healthcare Use Case:

  1. ALEX STARTS A NEW JOB: Alex starts a new job at Company A and provides the Human Resources team at his new company with a DID to be issued an employment credential with his job positions and further details.

    • TRANSACTIONS: Alex becomes an employee of Company A and is issued an employment credential.
    • DESIRABLE CORRELATION: Alex wants to be correlated as an employee of Company A so he can get access to pension plans, employees' benefits and so on.
    • CORRELATION THREAT:

  2. ALEX SIGNS UP FOR A HEALTHCARE SERVICE PROVIDER: Alex joins a new healthcare provider. Alex can get a discount as an employee of Company A, due to a partnership between his company and the healthcare service provider. Alex presents his employment credential to the provider with eluded data.

    • TRANSACTIONS: Alex uses his employee credential to prove he works at Company A but eludes any irrelevant personal information.
    • DESIRABLE CORRELATION: Alex wants to be correlated as an employee of Company A.
    • PRIVACY THREAT: Alex does not want to share any further personal information with his new healthcare provider.

  3. COMPANY A WANTS TO CHECK ON THE PARTNERSHIP WITH HEALTHCARE PROVIDER: Company A wants to check if the partnership with the healthcare provider is worth it to continue investing in. Company A requests data from the healthcare provider on the use of their employees.

    • TRANSACTIONS: Company A requests a data report from the healthcare provider.
    • DESIRABLE CORRELATION: Healthcare provider wants to correlate clients with employees of Company A.
    • CORRELATION THREAT: Healthcare provider does not want to accidentally reveal private information of their clients to Company A; Healthcare provider wants to only send data related to employees of Company A.

  4. HEALTHCARE PROVIDER SENDS INFORMATION BACK TO COMPANY A: In order to reduce correlation, the healthcare provider decides to send back only the number of credentials with eluding of the employee's name so they can be sure to preserve privacy.

    • TRANSACTIONS: Healthcare provider sends data back to Company A with the number of clients that are verified employees of Company A without showing their names.
    • DESIRABLE CORRELATION: Healthcare provider correlates clients to be employees of Company A.
    • ELISION BY HOLDER: Healthcare provider eludes the names of the clients.

Technology

In this section, we can develop a list of technical characteristics with regard to (anti-)correlation, and try to map how different technology choices can fulfill them.

TODO:

  • Maybe better differentiate between disclosure of data on various levels: credential, identifier, protocol, environment ("ambient correlation"?). On which layer does the (anti-)correlation happen, e.g. does it happen in the cryptographic signature, or on the identifier (DID) layer, etc.?
  • How does salting fit in? Who does the salting?
  • How do "sub-holders" fit in?

Characteristics:

  • [C1] The Holder can selectively disclose data with a Verifier.
  • [C2] The Holder can selectively disclose data with a Verifier, without requiring interaction with the Issuer during the disclosure process.
  • [C3] The Holder can selectively disclose data with different Verifiers, without requiring interaction with the Issuer during the disclosure process.
  • [C4] The Holder can arbitrarily choose which subset of data to selectively disclose with Verifiers, without requiring interaction with the Issuer during the disclosure process.
  • [C5] When disclosing data to multiple different Verifiers, the Verifiers cannot correlate the subject via any single, unique identifier.
  • [C6] When disclosing data to the same Verifier in multiple different interactions, the Verifier cannot correlate the subject via any single, unique identifier.
  • [C7] The Verifier can verify that the Holder who is disclosing the data is the same entity that has received the data from an Issuer. <-- Note: Maybe reference the work of the other RWoT11 group that's working on "holder binding"
  • [C8] The Holder can selectively disclose data that is derived (using predicates) from the original data they have received from an Issuer.

Choices:

Mapping of Characteristics and Technologies

Technology C1 C2 C3 C4 C5 C6 C7 C8
PLAINSIG ✕ ✕ ✕ ✕ ✕ ✕ ✓ ✕
SUBCREDS ✓ ✓ ✓ ✕ ✕ ✕ ✓ ✕
SINGLEUSECREDS ✓ ✕ ✕ ✕ ✓ ✓ ✕ ✕
BBS2020-LD ✓ ✓ ✓ ✓ ✕ ✕ ✓ ✕
BBS-SIG ✓ ✓ ✓ ✓ ✓ ✓ ✕ ✕
ANONCREDS ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓
SD-JWT ✓ ✓ ✓ ✓ ✕ ✕ ✓ ✕
JWP * ✓ ✓ ✓ (✓) (✓) (✓) ✓ (✓)
REDACTION2016-LD ✓ ✓ ✓ ✓ ✕ ✕ ✓ ✕
MERKLE2021-LD ✓ ✓ ✓ ✓ ✕ ✕ ✓ ✕
COCONUT ✓ ✓ ✓ ✓ ✓ ✓ ✕ ✓
GORDIAN ** ✓ ✓ ✓ ✓ (✓) ? (✓) ?

* C4, C5, C6, C8 depend on the underlying JSON Proof Algorithm (JPA).

** C5 can be done by issuers, but not holders. C7 can be achieved via encryption techniques.

Conclusion

TODO

Acknowledgements

We would like to acknowledge the following persons who contributed ideas that helped define the ideas in and the shape of this paper:

  • Adrian Gropper
  • Joe Andrieu

We also acknowledge the contributors to the 2019 paper:

  • Lionel Wolberger, Platin.io
  • Brent Zundel, Evernym/Sovrin
  • Zachary Larson, Independent
  • Irene Hernandez, Independent
  • Katryna Dow, Meeco
  • Christohper Allen, Blockchain Commons

References

Regulations that require some form of Data Minimization

  • The California Consumer Privacy Act (CCPA) in the United States, which requires businesses to disclose the categories of personal information that they collect, use, disclose, and sell and also to obtain consumer consent for the sale of personal information.
  • The Payment Card Industry Data Security Standard (PCI DSS), which is a set of security standards for organizations that handle credit card information and which requires merchants to limit the amount of cardholder data that is stored, processed, or transmitted.
  • The Personal Information Protection and Electronic Documents Act (PIPEDA) in Canada, which requires organizations to collect only the personal information that is necessary for the identified purpose.
  • The Health Insurance Portability and Accountability Act (HIPAA) in the United States, which requires healthcare organizations to protect the privacy of patient health information and to collect only the minimum necessary information to provide care.
  • The Cybersecurity Information Sharing Act (CISA) in the United States, which encourages organizations to share information about cybersecurity threats, but also requires companies to protect personal information and to limit the collection of data to that which is necessary to address the threat.

ISO standards related to data protection and privacy

  • ISO/IEC 27001:2013 - Information security management systems (ISMS) provides a framework for managing sensitive information and includes a requirement for organizations to minimize the amount of sensitive information that is collected, used, and retained.
  • ISO/IEC 29100:2011 - Privacy framework provides a framework for protecting personal information and includes a requirement for organizations to only collect personal information that is necessary for the specific purpose for which it is being collected.
  • ISO/IEC 27701:2019 - Extension to ISO/IEC 27001 and ISO/IEC 27002 for privacy information management provides a framework for managing personal information and includes a requirement for organizations to minimize the amount of personal information that is collected, used, and retained.
  • ISO/IEC 29151:2012 - Information technology - Security techniques - Data-at-rest protection provides guidance for protecting data when it is stored and includes a requirement for organizations to minimize the amount of sensitive information that is stored.
  • ISO/IEC 27040:2015 - Storage security provides guidance for protecting data when it is stored and includes a requirement for organizations to minimize the amount of sensitive information that is stored.

Related RWOT Advance Readings & Topics

Links to other related documents

Citations

Footnotes

  1. Identity Crisis: Clearer Identity Through Correlation ↩

  2. TBD self soveriegn identity principles ↩