Draft 2: 4-2-04 Identity and relationship management A discussion paper Mike Martin Centre for Social and Business Informatics, Institute for Policy and Practice, University of Newcastle upon Tyne mike-martin@btconnect.com Introduction One of the key evolutions in information systems architecture which has taken place in the last decade has been the shift in the location of integration components and facilities from middleware, behind or underneath applications, to the hub/portal which is in front of applications. This removes the assumption that the components or applications which are being integrated are necessarily parts of the same domain of ownership and enterprise and changes what we mean by "application". In this reconfiguration of the ICT infrastructure, the concept of the index as the means of establishing and maintaining consistency of identity and identification within a domain of integration and of relationship, has become a central issue. The purpose of this paper is to explore the issues of identity and identification at the integration and federation levels, that is to say within domains of mutual trust and understanding and between domains of trust and understanding. The current presenting problems in the real world of systems development are described, at the user level, in terms of issues such as “single log on”, authentication and identity management and at the computational level in terms of the handling of state among web services and resources. In the architectural view which will be presented here, the relationship between the efforts of initiatives such as Liberty Alliance (http://www.projectliberty.org/) and those of Globus Alliance (http://www.globus.org/) appear to be more than merely nominal even if these communities seem to have little to say to each other (so far). Proposals and developments in these particular initiatives have been pragmatic and have mostly been focused on immediate requirements in Net based commerce. When these ideas are applied in the world of public services, particularly when we include health and social care, we are faced with a more complex and subtle set of requirements compared with the relatively simple worlds of commercial product and service relationships. This paper argues that it is in the public service domain that we will find the real challenges for the next generations of information and communications infrastructures. If there are useful meanings to be constructed for terms such as "ambient intelligence" and "ubiquitous services", it is here that we will discover them. Defining some terms If we are to be able to talk about systems that deal with identities, there are a number of distinctions we must make and terms we must sort out. In particular we must be clear about the difference between the labels we attach to things and the things themselves. The former exist in our language and our information systems and are referred to, technically, as intensions. The latter exist in the world which is external to our systems and are called extensions. Consider the case of identifiers, such as my national insurance number and my NHS number, and me, the individual to whom they refer, who is employed in Great Britain and is sometimes a patient of the National Health Service. There were definite acts, performed in different organisational and administrative settings, that generated these particular identifiers, allocated them to me and recorded the fact in certain registers. Those systems were responsible for ensuring that, within each register or list of allocated identifiers: no two different individuals are given the same identifier, each individual has only one entry, 1 Draft 2: 4-2-04 that, within reasonable and appropriate constraints, the link between an individual and an identifier can be validated. The way that the last of these conditions is achieved is by associating a number of items of data, that are indicative of a specific individual, to each identifier in a register. Such things as given name, date and place of birth, the names of parents and current address are commonly used. Stronger and theoretically more reliable verification can be based on biometric data such as fingerprint, retinal scan or DNA profile. The purpose of this data is to support authentication processes, i.e. the validation of a claimed identity. Thus, we have a third concept which must be distinguished from both identifiers and the individuals to which they refer: an identity is a set of indicative data used to recognise and authenticate an individual. Notice that different registers may use different data items, called identity attributes, and any particular set of attributes may or may not provide, or continue to provide, a dependable basis for the authentication of a particular individual. Possible faults in an attribute data set used as an identity are: Under-indication: authentication data indicates more than one individual. Misindication: authentication data indicates the wrong individual. Non-indication: authentication data does not indicate anybody. The notion of individuality that applies to me is, of course, a physical one. There is only one of me and, as long as I exist, (in the period between the event of my birth and the event of my death) I can only be in one place at a time. There is another class of “individual” which has important characteristics concerning identity. The members of this class are legal entities such as companies, corporations or other collectives such as clubs and associations, which are constituted and operate under some public registration framework so as to be able to engage in both internal and external transactions and relationships. The ideas about identifiers and identities we have defined for individual people also apply to these virtual individuals but the external reality - the extension - that they refer to is constituted by a set of human laws and contracts rather than by the laws of logic and physics. So, my business partner and I own, and together, are a Company. This Company has a registered name and head office, a unique registration number on the British Companies' Register, a VAT number etc. We have defined the terms, identifier and identity together with the concept of a register which maintains the link between them and allows an identity to be associated with an individual. From the examples, we have shown that a particular individual may have a number of different identifiers and this leads to the requirement for indexing. An index is a (table) mechanism for associating identifiers which designate the same identity. The row of an index records the a set of identifiers from different relationship domains which are associated with a particular individual. The columns map the distribution of the clients of the provider domain. Returning to our example: if my NHS number and my National Insurance number appear on the same row of a shared index, a possibility for the clinical and the employment domains to communicate with each other about me has been created. Transaction and relationship. The reason why identity is important is because it is through identity that we can have relationships that persist over time and in different places. So identity and relationship are closely linked and the concept that ties them together is that of a transaction. The use of information from previous transactions in subsequent transactions, by the same parties, constitutes a relationship between them. The necessary conditions for a relationship are: the same parties engage in multiple encounters, they recognise each other, 2 Draft 2: 4-2-04 they remember and make use of their mutual history and, as a result they engage in further transactions. Anonymity Anonymity, like identity, is defined in relation to transaction. A party is anonymous if no identity related information regarding them is generated. A consequence of this is that an anonymous transaction can not be part of a relationship, there is no means by which the party can subsequently be recognised and any responsibility attached to them. A party is pseudonymous if there identity is entirely local to the immediate transaction or relationship and does not include indicative data which could be correlated with a domain in which it may be nested such as a public identity. Publication and permissioning Granting access to an index is publishing the fact that certain relationships exist between a set of relationship providers and a set subjects. (We use these terms to distinguish the column identifiers from the row identifiers.) R k, R Pb l, P R b m , R Pb a, R Pb b, R Pb c, R Pb d, Pb A relationship type + A provider identity Identifiers associated with the same individual. Figure 1: The implementation of an index as a table Access to an index generates the possibility of services which implement consents and permissions for transactions and relationships. The form of a permission could be the following: I, the individual with relationship W to individual I know as ABC, am willing to enter transactions e, f or g with anyone who has relationships X, Y or Z with ABC. Such publications may be subject initiated, subject permissioned, joint between subject and supplier or subject independent. Each case has consequent editorial rights, responsibilities and consequences. An individual taking advantage of this service can do so with the query: With whom can I engage in transaction e, regarding the individual I know as LMN? 3 The INDEX R k, R Pb l, R Pb m R , Pb a, R Pb b, R Pb c, R Pb d, Pb Draft 2: 4-2-04 A narrow cast channel of communication about a client. Identifier Identity attributes Record Provider of relationship Rb. Provider of relationship Rk. The client Figure 2: integrated relationships The INDEX R k, R Pb l, R Pb m R , Pb a, R Pb b, R Pb c R , Pb d, Pb The result of such a query will be a set of messages which are narrow-cast to the parties that share the indicated relationships with the subject and whose relationships have associated permissions – i,e, there local identifiers for this client are entered in the same row of the index. These messages will only identify the subject to each party my means of their local identifiers within their relationship and the process preserves pseudonymity. Any sharing of information about mutually identified subjects would be part of a subsequent local transaction between the collaborating parties. Identity Management Provider Provider of relationship Rb. Provider of relationship Rm. Figure 3: locally managed identity The relationships entered into a particular row of an index are considered integrated. Any provider who has placed a local client identifier on a row can narrowcast requests for transactions regarding that particular client to any other providers on that row. Identifiers referring to the same individual may be placed on different rows of the same index thus creating uncorrelated local identifiers. We now introduce a special type of relationship supplier, the identity manager, who operates in the following way: if a client who has identities entered on more than one row of the index forms a relationship with an identity manager, any narrow cast message on any of the rows occupied by this client will be transferred to the relationship providers on the other rows under a set of local rules and permissions. This is represented on Figure 3. 4 Draft 2: 4-2-04 Federated permissioning If we now introduce the concept of an identity manager who provides services to clients who have relationships in different index domains and who operates a further set of protocols and consents for routing client related narrow cast messages, we have a means of implementing a distributed and federal identity management infrastructure. This is represented in Figure 4. Registers which use different attribute sets to indicate identities. Register 1 Register 2 Register 3 IM P Ra b , Rb Pb ,P Rc b , Rd Pb , Re Pb ,P Rf b ,P Rg b ,P IM b Pc IM P Rk b ,P Rl b ,P R b m , Ra Pb , Rb Pb ,P Rc b , Rd Pb ,P IM b Pa Identity Management Provider B Identity Management Provider A Two domain indexes with shared identity management services Sets of profiles and records of the same individual with different relationships in two different domains. Relationship Rb. Relationship Rc. Relationship Rk. Relationship Ra. Fugure 4: Federated identity management In this scenario our client has two relationships in each of two index domains, i.e. domains of trust and service coordination. Relationships Ra and Rc are integrated and their providers are able to share information and transactions. Relationship Rk and Rb are coordinated via identity manager A who implements a set of agreed rules about which requests for which transactions are passed between them. The link between relationships Ra and c, on the one hand, and relationships Rk and b, on the other, are handled by federal identity management service B which links relationship Rk to Ra and Rc creating a transaction request routing between the two domains regarding this client. Note further that identifiers a and c are mutually traceable because they were originally authenticated through register 1. Relationships k and b are not reliably traceable to each other outside of the system because they are each based on authentications from different registers which use non identical attribute sets. Distribution and centralisation We have developed transactional concept of relationship and a relational concept of identity. We have further defined an identity management architecture based on these concepts which involves registration and indexing. With this framework we are now able to identify the different forms that distribution or centralization could take and, in this way, define the range of possible policies and consequent configurations of an identity management infrastructure. 5 Draft 2: 4-2-04 Our identity information architecture defines three levels: 1. The registration level, 2. The indexing level, 3. The relationship level. In the terms of our architecture, relationships at the third level are nested in collective relationships at the first level. At the level of a federation, policy could mandate a single register in an attempt to guarantee the ultimate traceability of all identities. Alternatively, the benefit of closer association between registers and their members, in terms of data quality and testability of coverage, may justify a policy of multiple, relatively local registers. Recursion here would imply an indexing of registers and the meta-registration of local registrars, producing an identity hierarchy rather than centralised identity. Policy might dictate that there is only one (logical) index and that all identity management is “local” to it. Such an approach has limited political stability requiring the exercise of strong central power. A further control option would be that individuals may only have identifiers on a single row within an index: all the relationships of a given individual are correlated and may be linked. While multiple indexes may be sanctioned, policy could dictate that there is only one, universal identity management service which is present in all of them. This places very high demands on the trustworthiness and dependability of such a provider. Alternatively, there could be a central licensing authority so that identity management service providers are registered and regulated. Finally, there could be a free market in identity service providers. Such policies, of course, have economic and commercial consequences as well as political ones. Concluding remarks Current developments and, in particular, Liberty Alliance, represent a pragmatic and evolutionary approach to the issues of identity in the context of commercial transactions and relationships and current technology and systems practice. As such, we believe that it is a specialization of the framework presented here. Current developments, which respond to commercial and infrastructural realities, tend to conflate the idea of the authentication role of the register and the transactional support of the index into a single third party system onto which the concept of the circle of trust and a peremissioned identity management service can be mapped. Hopefully, the most useful way of regarding the architecture presented here is as an attempt to represent and rationalise the current and expected evolution of identity management concepts taking into account the different requirements of public service, administration and commerce. 6