This is the first in a series of posts that cover the Field Guide to Internet Trust Models Paper. The paper was presented at the University of Texas at Austin ID360 Conference in 2013.
This paper was collaboration between myself and Steve Greenberg. I had an outline of all the Trust Models and worked with Steve Greenberg for several months to shape it into the paper.
The full papers is downloadable TrustModelFieldGuideFinal (see the bottom of this post for a link to a post on each of the models).
The decreasing cost of computation and communication has made it easier than ever before to be a service provider, and has also made those services available to a broader range of consumers. New services are being created faster than anyone can manage or even track, and new devices are being connected at a blistering rate.
In order to manage the complexity, we need to be able to delegate the decisions to trustable systems. We need specialists to write the rules for their own areas and auditors to verify that the rules are being followed.
This paper describes some of the common patterns in internet trust and discuss some of the ways that they point to an interoperable future where people are in greater control of their data. Each model offers a distinct set of advantages and disadvantages, and choosing the appropriate one will help you manage risk while providing the most services.
For each, we use a few, broad questions to focus the discussion:
- How easy is it for new participants to join? (Internet Scale)
- What mechanisms does this system use to manage risk? (Security)
- How much information the participants require from one another how strongly verified?
(Level of Assurance -not what I think assurance is…but we can talk – it often also refers to the strength of security like number of factors of authentication )
Using the “T” Word
Like “privacy”, “security”, or “love”, the words “trust” and “identity”, and “scale” carry so much meaning that any useful discussion has to begin with a note about how we’re using the words.
This lets each link the others to past behavior and, hopefully, predict future actions. The very notion of trust acknowledges that there is some risk in any transaction (if there’s no risk, I don’t need to trust you) and we define trust roughly as:
The willingness to allow someone else to make decisions on your behalf, based on the belief that your interests will not be harmed.
The requester trusts that the service provider will fulfill their request. The service provider trusts that the user won’t abuse their privileges, or will pay some agreed amount for the service. Given this limited definition, identity allows the actors to place one another into context.
Trust is contextual. Doctors routinely decide on behalf of their patients that the benefits of some medication outweigh the potential side effects, or even that some part of their body should be removed. These activities could be extremely risky for the patient, and require confidence in the decisions of both the individual doctor and the overall system of medicine and science. That trust doesn’t cross contexts to other risky activities. Permission to prescribe medication doesn’t also grant doctors the ability to fly a passenger airplane or operate a nuclear reactor.
Trust is directional. Each party’s trust decisions are independent, and are grounded in the identities that they provide to one another.
Trust is not symmetric. For example, a patient who allows a doctor to remove part of their body should not expect to be able to remove parts of the doctor’s body in return. To the contrary, a patient who attempts to act in this way would likely face legal sanction.
Services and APIs change faster than anyone can manage or even track. Dealing with this pace of change requires a new set of strategies and tools.
The general use of the term “Internet Scale” means the ability to process a high volume of transactions. This is an important consideration, but we believe that there is another aspect to consider. The global, distributed nature of the internet means that scale must also include the ease with which the system can absorb new participants. Can a participant join by clicking “Accept”, or must they negotiate a custom agreement?
In order to make this new world of user controlled data possible, we must move from a model broad, monolithic agreements to smaller, specialized agreements that integrate with one another and can be updated independently.
A Tour of the Trust Models
The most straightforward identity model, the sole source, is best suited for environments where the data is very valuable or it is technically difficult for service providers to communicate with one another. In this situation, a service provider issues identity credentials to everyone it interacts with and does not recognize identities issued by anyone else. Enterprises employing employees, financial institutions, medical providers, and professional certifying organizations are commonly sole sources. Because this is the most straightforward model to implement, it is also the most common.
Two sole sources might decide that it’s worthwhile to allow their users to exchange information with one another. In order to do so, they negotiate a specific agreement that covers only the two of them. This is called a Pairwise Agreement and, while it allows the two parties to access confidential resources, the need for a custom agreement makes it difficult to scale the number of participants. This is also a kind of federated identity model, which simply means that a service accepts an identity that is managed someplace else.
As communication technology became more broadly available, the number of institutions who wanted to communicate with one another also increased. Groups of similar organizations still wanted to issue their own identities, but wanted their users to be able to interact freely with one another. The prospect of each service having to negotiate a custom agreement with every other service was daunting, so similarly chartered institutions came up with standard contracts that allow any two members to interact. These groups are called Federations, and there are several different kinds. Federation agreements and membership are managed by a Contract Hub.
When the federation agreement limits itself to policy, governance, and common roles, but leaves technical decisions to the individual members, it’s referred to as a Mesh Federations. Individual members communicate form a mesh, and can communicate directly with one another using whatever technology they prefer.
Alternatively, a Technical Federation defines communication methods and protocols, but leaves specific governance and policy agreements to the members. In some cases, the technical federation may also route messages between the members.
As the number of services has increased, so has the problem of managing all of those usernames and passwords. Users might decide to reuse an existing identity rather than creating a new one. In recent years, some organizations have made identities that they issue available to other services. Service providers accept these identities because it lowers the cost of user acquisition. When the same entity provides identities for both the requester and the service provider, it is referred to as a Three Party Model.
If the requester and the service provider have provider have separate but compatible identity providers, it is called a Four Party model. This is present in highly dynamic models, such as credit card processing,
Peer-to-peer networks are for independent entities who want to identity assurance, but who lack a central service that can issue identities to everyone. To get around this, the participants vouch for one another’s identities.
Individual contract wrappers are an innovation to enable complex connections between services where the terms and conditions of using the data are linked to the data.
Common Internet Trust Models
Sole source: A service provider only trusts identities that it has issued.
Pairwise Federation: Two organizations negotiate a specific agreement to trust identities issued by one another.
Peer-to-Peer: In the absence of any broader agreement, individuals authenticate and trust one another.
Three-Party Model: A common third party provides identities to both the requester and the service provider so that they can trust one another.
“Bring your Own” Portable Identity: In the absence of any institutional agreement, service providers accept individual, user-asserted identities.
“Winner Take All” Three Party Model: Service provider wants to allow the requester to use an existing identity, but only accepts authentication from a single or very limited set of providers.
Federations: A single, standard contract defines a limited set of roles and technologies, allowing similar types of institution to trust identities issued by one another.
Mesh Federations: These share a common legal agreement at the contract that creates permissible interoperability.
Technical Federations: These share a common technical hub responsible for making the interoperability happen.
Inter-Federation Federations: This is what happens when one federation actually inter-operates with another federation.
Four-Party Model: An interlocking, comprehensive set of contracts allows different types of entity to trust one another for particular types of transaction.
Centralized Token Issuance, Distributed Enrollment: A shared, central authority issues a high-trust communication token. Each service provider independently verifies and authorizes the identity, but trusts the token to authenticate messages.
Individual Contract Wrappers: Manage how personal data is used rather than trying to control collection. Information is paired contract terms that governs how it can be used. Compliance is held accountable using contract law.
Open Trust Framework Listing: An open marketplace for listing diverse trust frameworks and approved assessors.