Peng Liu
School of Information Sciences and Technology, The Pennsylvania State University, 003B Thomas Building, University Park, PA 16802. E-mail: pliu@ist.psu.edu
Amit Chetal
Department of Computer Science and Engineering, The Pennsylvania State University, 220 Pond Laboratory, University Park, PA 16802. E-mail: chetal@cse.psu.edu
The September 11 attack and the following investigations show that there is a serious information sharing problem among the relevant federal government agencies, and the problem can cause substantial deficiencies in terrorism attack detection. This paper performs a systematic analysis of the causes of this problem; and it finds that existing secure information sharing technologies and protocols cannot provide enough incentives for government agencies to share information with each other without worrying that their own interests can be jeopardized. Although trust-based information access is well studied in the literature, the existing trust models, which are based on certified attributes, cannot support effective information sharing among government agencies, which requires an interest-based trust model. To solve this information sharing problem, this paper proposes an innovative interest-based trust model and a novel information sharing protocol, where a family of information sharing policies are integrated, and information exchange and trust negotiation are interleaved with and interdependent upon each other. In addition, an implementation of this protocol is presented using the emerging technology of XML Web Services. The implementation is totally compatible with the Federal Enterprise Architecture reference models and can be directly integrated into existing E-Government systems.
1. INTRODUCTION
1.1. Information Sharing Problems among Government Agencies
For one, agencies acted as stovepipes, or rigid functionally organized departments [71]. For instance, there were multiple watch lists that existed across the federal agencies in regards to suspecting malicious terrorists. Each agency had a separate watch list based on separate data organization and schema, but none of the watch lists had details about the other and the watch lists were not unified by any means. Moreover, there was a lack of communication between agencies. In fact, even President Bush conceded, “In terms of whether or not the FBI and the CIA were communicating properly, I think it is clear that they weren’t” [27]. So, the government agencies could not aggregate their information properly.
1.2. Development of E-Government Initiatives
Because of the stovepipes that existed between several government agencies, the president and the rest of the congressional staff developed 24 E-Government initiatives [36]. These initiatives could generate several billion dollars in savings and specify plans so agencies can collaborate and share their information in a more efficient manner, thereby consolidating information.
1.3. Establishing the Federal Enterprise Architecture
After the terrorist attacks of September the 11th on American soil, much of the nation spoke out for a drastic change within the federal government. One of the primary changes focused on addressing information sharing amongst government agencies. With efficient information sharing, government agents will be able to predict and possibly preempt attacks form occurring. However, prior to September the 11th, serious information sharing problems existed within government agencies.
In order to address and fulfill the E-Government initiatives, there needs to be a flexible, comprehensible, standardized (reference) model. This model is known as the Federal Enterprise Architecture (FEA) [31]. It is vitally important for agencies to communicate in a more citizen-centric, unifying way, where information can be exchanged in a more efficient manner. As a result, this new FEA will leverage technology investments to avoid unnecessary duplication of infrastructure amongst agencies, link business processes through shared, yet protected information systems, and leverage disparate
1
business processes [7]. It is clear that enabling effective information sharing is a major goal of the FEA. (In effective information sharing, the information held by one agency, if useful to another agency, will be timely disclosed to the other agency in a cooperative, proactive manner in most, if not all, cases.) In fact, one of the most intriguing E-Government initiatives is the E-Authentication initiative, whose objective is: “Build and enable the mutual trust needed to support widespread use of electronic interactions between the public and management requirements for effective information sharing among government agencies. Second, this paper proposes an innovative interest-based trust model and a novel information sharing protocol, where a family of information sharing policies are integrated, and information exchange and trust negotiation are interleaved with and interdependent upon each other. Third, an implementation of this protocol is presented using the emerging technology of XML Web Services. The implementation is totally compatible with the FEA government and across governments…” [36].
1.4. Limitations of Existing Information Sharing Technologies
However, existing information technologies are very limited to achieve the goal of the E-Authentication initiative, i.e., building mutual trust between agencies in such a way that effective information sharing can be enabled, since they cannot build the level of trust that is needed to provide enough incentives for government agencies to share information with each other without worrying that their own interests can be jeopardized. Although the FEA provides a solid, standardized platform for agencies to collaborate with each other, the FEA assumes the use of existing trust models, and agencies are still reluctant to share sensitive information with one another. Several government agencies declared that they were reluctant to share sensitive information with one another and are not comfortable since they don’t entirely trust the other agencies. This study shows that although trust-based information access is well studied in the literature [11, 12, 17, 18, 44], the existing trust models, which are based on certified attributes, cannot support effective information sharing among government agencies, which requires an interest-based trust model. Information sharing among government agencies requires a different, more restrictive trust model primarily due to two reasons: (1) much of the information that each agency carries is highly sensitive, so agencies find it difficult to simply let another agency gain access or observe its sensitive information without first developing enough trust with the other agency; (2) the differences and conflicts-of-interest existing between agencies require a more accountable and fair information sharing procedure, where the (mutual) trust is not only affected by certified attributes and cross-agency authority-mapping, but also affected by whether a win-win information sharing can be achieved without jeopardizing the interests of any agency involved.
1.5. Our Contributions
First, this paper performs a systematic analysis of the causes of the information sharing problem among government agencies, and it identifies the unique trust
2
reference models and can be directly integrated into existing E-Government systems. We believe the proposed trust model and information sharing protocol may dramatically improve the effectiveness of information sharing among government agencies and reduce the deficiencies in countering terrorism attacks. 1.6. Paper Organization
The rest of the paper is organized as follows. In Section 2, we address the FEA and its limitations. Section 3 addresses “which kind of trust models is necessary to enable effective information sharing among government agencies?” In Section 4, we survey the literature of information sharing and show why existing information sharing technologies cannot support effective information sharing among government agencies. Section 5 presents our trust model and our information sharing protocol. In Section 6, we present an XML Web Services based implantation of our information sharing protocol. In Section 7, we conclude the paper.
2. FEDERAL ENTERPRISE ARCHITECTURE
In this section, we provide an overview of the FEA so that in Section 3 we can show why the FEA is limited in enabling effective information sharing, and in Section 6 we can present a FEA-based implementation of our information sharing protocol.
The FEA exists as a function-oriented framework for describing the business operations of the federal government. It is also independent of the agencies that perform those operations [35]; this means that the architecture provides support for all agencies, independent of each agency’s operations.
2.1. Principles of Federal Enterprise Architecture
Initially, it is important to understand that there are several principles that form the basis for the federal enterprise architecture [35]. For one, since much of the information stored by each agency requires proprietary software dependencies, there should be an establishment of standards, mainly the establishment of federal
interoperability standards. By using the emerging new web standards such as XML and XML Web services, agencies will be free from software and hardware dependences. These new standards will allow for increased collaboration and cross-agency information exchanging. Also, another vital principle involves capitalizing on standardization measures based on common functions between agencies. Federal agencies are attempting to develop reusable functions that exist across agencies and purchase architecture components that will allow for increased collaboration and eliminate redundancy [31].
Moreover, another principle revolves around coordinating the technology investments among the federal agencies. Since there is constant overlap of functions between the different agencies, vertical and horizontal integration will allow for a reduction in spending, causing a huge increase in federal financial savings.
Finally, a critical principle is to protect federal information against malicious attacks and unauthorized access. Since much of the federal information will be represented in a more collaborative manner through new web standards such as XML and web services [10], new security measures need to be taken. Mainly, sensitive information is no longer designated to one agency but rather is flowing among agencies, so sharing such sensitive information needs to be better protected because the damage caused by an attack can be dramatically magnified across multiple government agencies through the “bridges” built by the FEA.
2.2. Models of Federal Enterprise Architecture
The principles that make up the federal enterprise architecture clearly address the existing stovepipes that exist between agencies and address the issues behind the E-Government initiatives. However, to have a deeper understanding of the FEA, we need to discuss the four reference models of the FEA, namely the business reference model, the data reference model, the application capability reference model, and the technical reference model [33]. The four models are connected in a hierarchical fashion in the sense that each lower level model is a detailed layout (or exploration) of typically one aspect of the corresponding higher level model. These models depict the process of how the federal enterprise architecture is used for cross-agency information sharing and collaboration.
2.2.1. Business Reference Model
First, the business reference model describes the lines of business of each agency, and identifies the customers and partners of each agency. The business reference model is a high level view of the federal enterprise architecture
3
based on a hierarchical construct for describing the business operations of the federal government [33].
The basic procedure (to develop this reference model) involves looking at previous efforts that agencies used to identify their functions, and then once the list was formed in its entirety, categorize the functions into designated mutually exclusive business areas.
2.2.2. Data Reference Model
The data reference model shows how the components of the business reference model can be used to develop cross-agency information exchanges, and it provides a form of data standardization between the agencies. The data reference model will be used to describe the types of interactions and information exchanges that occur between the federal government and its various customers. XML, extensible markup language, plays a critical role in developing the E-Government data reference model. XML provides a mechanism for federal lines of business to define and standardize XML schemas so some lines of businesses can interact with other lines of businesses. XML provides a standard way for preserving and communicating information encoding, tagging, and internationalizing information [73]. In XML, tags can be created to represent each function for a particular line of business. Also, these XML tags can be transmitted via HTTP or some other protocol so other agencies can then use these common functions, create new ones, or build upon them.
2.2.3. Application Capability Reference Model
The application capability reference model describes how the application capabilities support the business objectives [16]. The application capability reference model is comprised of two sub-models: the conceptual process model and the interoperability model.
The conceptual process model, as shown in Figure 1, describes the bridge between the functional view of the business reference model and the technology needed to carry out the cross-agency information exchanges. In order to retrieve information from an agency application, six layers of services (i.e., the end user layer, the access portal layer, the crosscutting requirements layer, the web platform layer, the applications interface layer, and the enterprise data and applications layer) need to be developed so that the information exchanges can take place and each agency will be able to receive his/her desired information. The interoperability model, as shown in Figure 2, describes the primary application components that support the conceptual process model and how they interoperate within and across the lines of businesses [31].
The interoperability occurs at the user, data, and application level.
FIG. 1. Conceptual Process Model FIG. 2. Interoperability Model 2.2.4. Technical Reference Model Finally, the technical reference model, as shown in Figure 3, shows how technology is being used to deliver
application capabilities. The technical reference model defines how the hardware, software, and physical location
can support the businesses, data, and functions [61]. The
purpose of the technical reference model is for better coordination, development, and support of the E-Government initiatives and cross-agency information sharing.
4
FIG. 3. Technical Reference Model
3. NEED FOR AN INTEREST-BASED TRUST MODEL
In this section, before we show why FEA is limited in enabling effective information sharing among government agencies, we will first identify the agencies’ specific requirements on the (mutual) trust model.
3.1. Why There Is a Lack of Trust among Agencies?
The primary reason that spurs the hesitation and reluctance for agencies to share their sensitive information with one another is that they simply don’t trust each other. The lack of trust is due to several reasons. We believe a major reason is that conflicts of interest usually exist
between agencies and as a result, having agency A share
some information with agency B may actually compromise the interests of agency A. In fact, “the conflict between the agencies is deeply rooted. Even an event as horrible as [Sept 11] cannot erase five decades of turf battles” [40]. Agencies may experience a great deal of friction in cooperating in a joint effort, and information sharing usually plays an important role in these frictions. As a result, unfair information sharing and misuse of shared information can substantially discourage the agencies involved in a joint effort to share information with each other. To illustrate, when two agencies A and B are
cooperating in a joint effort, if agency A shares a piece of
information (with B) which is critical to a task assigned to B, but B hides a piece of information which is critical to a
task assigned to A, then B can perform much better than A, and the performance difference may give B a lot of
advantages in competing with A. Hence, to ensure that his/her interests will not be jeopardized, agency A may
want to do the same thing as B does, that is, A will stop sharing information with B (proactively). In this way, as more agencies hide their information, there will be less and
less mutual trust staying between agencies. Moreover, the above problem can be further exacerbated by two facts: (a) there are a lot of misunderstandings between government agencies; (b) little accountability is provided in existing information sharing processes. Fact (a) indicates that agencies may possess substantial misunderstandings (or doubt) about one another in terms of the intent, objective, and strategies of information sharing due to some inherent differences in their agency structures, cultures, intra-agency policies, beliefs, responsibilities, and missions. For example, “the deeply ingrained differences between the CIA and FBI have long prevented them from working together...” [40]. The misunderstandings between agencies can make one agency “sense” more risk or less trust when sharing information with another agency, thus they can further discourage agencies to share information.
Fact (b) exacerbates the above problem since the lack of accountability makes it difficult to resolve the conflicts caused by unfair or misused information sharing among agencies. Since the “bad” player cannot be easily identified or published, one agency will probably “sense” more risk or less trust when sharing information with another agency.
Another reason for the lack of trust between agencies is that shared information may be improperly processed by an agency due to the following observations. First, internal corruption might exist between government agencies. In fact, in a recent survey by the Computer Institute and the FBI, it was reported that a great deal of organizations face insider attacks [48]. Internal corruption within an agency may seriously jeopardize the other agencies’ trust in the agency. Second, responsibility and seriousness needs to take charge in order for cross-agency information sharing to take place. In fact, one Colonel Steve York, stated that, “The government has to lead by example. Within the government, agencies don’t trust each other. The government has to break down these walls …” [75].
3.2. Requirements on the Trust Model
Information sharing schemes are trust-based; and whether an information sharing scheme can lead to effective information sharing among government agencies is heavily dependent upon the trust model on top of which the information sharing scheme is constructed. The reasons about why there is a lack of trust among agencies imply that the trust model for effective cross-agency information sharing needs to satisfy the following requirements:
A. The trust model should be built in such a way that
win-win information sharing will always increase the mutual trust. An information sharing procedure between two agencies is a win-win procedure if the interests (or payoffs) of both agencies will be increased (to a similar degree) when the procedure ends. If win-win information sharing does not increase the mutual trust or even decreases the mutual trust, then the success of a win-win information sharing may not be able to promote more valuable and successful information sharing, since a higher level of mutual trust is usually needed to make the two agencies willing to share more sensitive or important information. On the other hand, if this requirement is satisfied, a win-win information sharing procedure can increase the mutual trust which can in turn encourage more valuable win-win information sharing, and since more sensitive information can be shared, more interests can be obtained by both agencies. This property has the potential to transform the situation where “nobody wants to share information” to the
5
situation where “everybody wants to share information”.
B. The trust model should be built in such a way that
win-lose information sharing will always decrease the trust of the loser in the winner. An information sharing procedure is a win-lose procedure if the interests of (at least) one agency will be decreased when the procedure ends. When requirement A is satisfied, this requirement implies that “bad” agencies that hide information to “steal” more interests can actually be punished due to the following reason: (a) win-lose information sharing hurts mutual trust; and degraded trust will prevent more information sharing; (b) hence, good agencies will have less and less win-lose information sharing with bad agencies and more and more win-win information sharing will good agencies; and as a result, after a relatively long period of time, all the good agencies will gain a lot of interests but the bad agencies can only gain very little. C. Requirements A and B indicate that interests must
be part of the trust model.
D. Requirement C indicates that information validity
and utility must be part of the trust model. This is because the interests that an agency can gain through an information sharing procedure are ultimately determined by the validity and utility of the information the agency received. A piece of information must be valid (i.e., with high integrity) in order to be useful. And the utility of the piece of information is also determined by such factors as whether it is needed by the agency and whether it is critical to solve a critical problem faced by the agency. 3.3. Limitations of Existing Trust Models
Of course, the trust model for information sharing among government agencies is built on the shoulder of existing trust models, which as we will shown in Section 4 shortly, can be broken down into three layers: identification and authentication build the first layer of trust; certified attributes built the second layer of trust; and delegation builds the third layer of trust.
However, as we will shown in Section 4 shortly, a serious drawback of existing information sharing technologies is that the existing trust models do not satisfy the four requirements we have identified above. Existing trust models can enable agencies to trust “who you are”, “the attributes you have”, “the authorities you have”, but cannot enable agencies to trust “this information sharing procedure will increase my interests” or “this information sharing procedure will be a win-win one”. In summary, credentials and authorities are the basis of existing trust models, but they cannot deliver interests guarantees,
however, information sharing among government agencies requires an interest-based trust model.
3.4. Limitations of FEA
Although the federal enterprise architecture provides a solid foundation for cross-agency information exchanging, it assumes the use of existing trust models which are not interest-based as we will show in Section 4 shortly. As a result, without a new interest-based trust model, even if the FEA is fully implemented, agencies will be still reluctant to freely share information with one another.
4. EXISTING INFORMATION SHARING TECHNOLOGIES
In this section, we present a classification of existing information sharing technologies and show why they are limited in enabling effective information sharing among government agencies. Existing information sharing technologies can be classified into two categories: (a) privacy-preserving information sharing, where two parties with information x and y respectively share their information with each other in such a way that a function of x and y, denoted f(x,y), is computed and learned by the two parties, but the privacy of x and y is preserved during the information sharing process, that is, the two parties learn only f(x,y) and nothing else; and (b) non-privacy-preserving information sharing, where two parties with information x and y respectively can not only share a function of x and y with each other but also share (part of) x and/or (part of) y with each other.
4.1. Privacy-Preserving Information Sharing
Privacy-preserving information sharing techniques can be broken down into three sub-categories: (a1) trust third party techniques; (a2) secure multi-party computation; and (a3) application specific techniques.
4.1.1. Trust Third Party The two parties give their data (i.e., x and y) to a
“trusted” third party and have the third party do the computation (i.e., computing the value of f(x,y)) [3, 49].
However, the third party has to be completely trusted, both
with respect to intent and competence against security breaches. The level of trust required may be too high for this solution to be practically used by government agencies. 4.1.2. Secure Multi-Party Computation
There are two parties with inputs x and y respectively (each one is typically a n-dimension vector) and the goal of
6
secure multi-party computation is to compute a function f(x,y) without involving a third party such that the two parties learn only f(x,y) and nothing else. See [42, 57] for a discussion of various approaches to this problem.
Yao [74] showed that any multi-party computation can be solved by building a combinational circuit, and simulating that circuit. A variant of Yao’s protocol is presented in [58] where the number of oblivious transfers is proportional to the number of inputs and not the size of the circuit. However, this approach is too expensive, especially in communication cost. For example, for n = 1 million, the communication time for the circuit-based protocol is 144 days (using a T1 line), which is clearly too long for government agencies to share information with each other.
4.1.3. Application Specific Solutions
Without involving a third party, efficient privacy-preserving information sharing is possible in some specific information sharing settings, where some specific data structures and some specific forms of secure multi-party computation (i.e., specific forms of f(x,y)) can be exploited substantially to reduce the complexity. For example[1], a researcher proposes a set of efficient protocols for information sharing across private databases, where the notion of minimal information sharing across private databases is formalized, and the unique characteristics of the relational data model and the associated query operators are exploited to achieve both privacy and efficiency. Compared with [74], for n = 1 million, the communication time of [1] is only 0.5 hours instead of 144 days.
4.2. Non-Privacy-Preserving Information Sharing Privacy preserving is actually not a requirement for most, if not all, information sharing activities between government agencies. The techniques for non-privacy-preserving information sharing can be broken down into
three sub-categories: (b1) privilege-based information
sharing; (b3) trust-based information sharing; and (b3) fair
data exchange.
4.2.1. Privilege-Based Information Sharing Access control technologies can be used to support both intra-organization and inter-organization information sharing. Within a single organization, information (stored in files or databases) can be shared by assigning proper
authorizations (or privileges) to the subjects (i.e., users). For example, when Alice and Bob are both authorized to access a piece of information x, x is shared by Alice and Bob. Authorization management is well studied in the field
O would like to offer (or share) a free software package only to college students, while O does not trust any university directly. Instead, O accepts accrediting credentials issued by the software Accrediting Board for Universities (ABU). As a result, to get the software for free, a student Alice of university StateU needs to show O not only a credential issued by StateU which says that of access control. For example, a popular authorization management scheme is role based access control [21]. Inter-organization access control focuses on how to allow a user U of organization A to access a piece of information x owned by organization B. To make the access control policy of each organization intact, inter-organization access control needs to map (or translate) U’s authorizations in organization A to some corresponding authorizations in organization B. Good methods to do such mappings are developed in the literature [4, 63]. However, inter-organization access control is limited in sharing information among government agencies, since inter-organization access control assumes that organizations trust each other with respect to both the validity of information and the validity of authorizations; however, this level of trust may not be “reachable” between two real world agencies in many information sharing scenarios.
Finally, for a set of organizations that distribute, store, and utilize their information in a decentralized fashion, digital credentials (on top of a PKI) can be used to facilitate the enforcement of both intra-organization and inter-organization access control. Using digital credentials, each organization can certify the roles, attributes, and authorizations associated with her employees, and the PKI can ensure that the digital credentials issued by each organization are verifiable so that forged credentials will never be used. In addition, to make such enforcements more efficient, privileges can be delegated from one party to another party [2, 41].
4.2.2. Trust-Based Information Sharing Although privilege-based information sharing may work well under centralized trust, privilege-based information sharing cannot handle information sharing under decentralized trust. Under centralized trust, there is one authority which everybody trusts, however, such authority does not exist under decentralized trust where one party typically trusts only a couple of other parties. We can use information sharing under centralized trust for trust-based information sharing. Trust plays an implicit role in privilege-based information sharing where every
credential can be verified in a straightforward way. In
contrast, trust plays a much more explicit role in trusted-based information sharing. For example, if organization A
does not trust organization B, A will not trust credentials
issued by B, and no employee of B can access information
owned by A unless he or she holds a credential issued by a
party trusted by A.
In trust-based information sharing, everybody
determines whether to share a piece of information (with
others) based on trust; and the access control is typically
enforced based on the attributes or roles associated with
the subjects. To illustrate, assume an online software store
7
Alice is a student of StateU, but also a credential issued by ABU which says that StateU is an accredited university of ABU. In other words, O establishes its trust in Alice through a chain of two credentials: the credential issued by ABU makes O trust StateU and its credentials, and the credential issued by StateU makes O trust Alice.
In summary, trust-based information sharing is composed of two key technologies: (1) trust management; and (2) attributed or role-based access control. Trust management enables a party to trust a credential (and its content) issued by a stranger through a chain of other “proving” credentials. Then attributed (or role) based access control can be used to disclose a piece of information based on the attributes certified by a trusted credential. Delegation is an inherent part of trust based information sharing. For example, in the above example, O in fact delegates the authority over the identification of preferred customers to ABU.
Trust management technologies can be broken down into three sub-categories: (1) credential-chain-based trust management; (2) trust negotiation; and (3) ad hoc trust management.
• Credential-chain-based trust management: In this
category, a chain of credentials issued by a set of
parties can be used to make one party trust another party with respect to the other party’s attributes. Technologies in this category include but are not limited to SPKI/SDSI [18], KeyNote [12], Policy-Maker [11], REFEREE [17], and Delegation Logic [44]. Moreover, in [51, 52], the concept of partial trust is introduced, where one may consider someone to be completely trusted, completely un-trusted, completely unknown, or somewhere in between. • Trust negotiation [46]: In this category, each party
can have multiple credentials containing different sets
of attributes. To establish trust between two parties,
they use access control policies that specify what
combinations of digital credentials a stranger must
disclose to gain access to a local information resource.
However, to preserve the privacy of sensitive
credentials, a credential will not be disclosed to party
B by party A unless party A has a certain level of trust
in B, which is established based on a set of credentials
disclosed from B to A. Therefore, A and B need to
interactively negotiate their trust on each other
through typically multiple rounds of trust-based
credential exchange. A party can use many different strategies to negotiate trust, however, to preserve parties’ autonomy, each party should ideally be able to choose its negotiation strategy independently, while still being guaranteed that negotiations will succeed whenever possible, i.e., that the two parties’ strategies will interoperate.
• Ah hoc trust management: In this category,
credentials are not necessary to establish trust. For example, in [45], a simple distributed trust model is proposed where numerical ad hoc trust levels are maintained for each party to measure the degrees to which it trusts the other parties. In [23], a protocol for maintaining and exchanging reputations in peer-to-peer networks is proposed. Note that reputations can be directly mapped to trust levels. In [55], a reputation-based trust model for peer-to-peer e-commerce communities is proposed.
4.2.3. Fair Data Exchange
Experience with e-commerce has shown that an exchange of one data item for another between mutually distrusted parties is usually the crux of an electronic transaction. Fair data exchange technologies have been used in many applications such as non-repudiation of message transmission [43], certified mail [24], contract signing [9], and e-payment systems [20, 70]. An exchange is fair if at the end of the exchange, either each party receives the item it expects or neither party receives any additional information about the other’s item. Fair data exchange protocols in the literature can be broken into two categories: third party protocols which use a trusted or semi-trusted third party and gradual exchange protocols [9] where the probability of correctness is gradually increased over several rounds of communications. Third party protocols can be further classified into two sub-categories: exchanges with online third parties [20, 39], where an exchange cannot be completed without using the trusted channel even if both parties play honestly, and exchange with offline third parties [5, 8], where an exchange can be completed without interferences of the trusted third party if the two parties play honestly and the third party is needed only when a party does not play honestly. A third party is trusted if it will neither misbehave on its own, nor conspire with either of the players. A third party is semi-trusted if the third party may misbehave on its own but will not conspire with either of the two parties [39].
Although information sharing usually involves data exchange, the type of information sharing in fair data exchange and the type of information sharing in E-Government are very different. In particular, first,
8
information sharing in fair data exchange is fairness-based, but information sharing in E-Government is trust-based. Fair data exchange protocols focus on fairness, non-repudiation and atomicity, but information sharing in E-Government focuses on trust, access control, and privacy. Trust is easy to establish in fair data exchange, but hard in information sharing in E-Government, where fairness is not necessarily a requirement. Information sharing in E-Government typically requires the two pieces of information to be exchanged between two parties are of similar types, but fair data exchange protocols typically exchange very different types of information.
As a result, fair data exchange technologies are limited in supporting information sharing among government agencies. Relying on a trusted third party may be too strong a requirement for practical information sharing. Using a semi-trusted third party cannot help much, since although the third party will not conspire with either of the two agencies, none of the two agencies may want to reveal the information they want to share with each other to the third party due to the sensitivity of the information. Finally, using a gradual data exchange protocol may be too expensive and cause substantial delays.
4.3. The Uniqueness of Our Trust-based Information Sharing
Scheme
Although our information sharing scheme is built on top of the set of existing information sharing technologies, it has several unique features. In particular,
• Our scheme is FEA specific. Our scheme is
motivated by a systematic analysis of the specific security requirements for information sharing in E-Government. To our best knowledge, our scheme is the only information sharing scheme that satisfies the set of security requirements identified in Section 3.
• Our scheme makes validity and utility of
information part of the trust model. That is, whether Agency A trusts agency B is also affected by the information (and its validity and utility) that A has obtained from B besides credentials. In credential-chain-based trust management and trust negotiation, trust is only affected by credentials. However, experience with real world cross-agency information sharing shows that even if Agency A believes that an agent of B has a set of eligible attributes, A may not want to share a piece of sensitive information with the agent due to some conflicts of interests between A and B.
• Interleaved information sharing and trust
negotiation. In [47], each information unit is only
associated with a combination of credentials, and the trust level is increased only due to the disclosure of more credentials. By contrast, in our scheme, the trust level may also be increased when the shared information is valid and appreciated. In [47], an information unit will be disclosed only when the trust negotiation succeeds. Otherwise, no information will be disclosed. By contrast, in our protocol, information sharing and trust negotiation are interleaved. Partial information sharing is in fact used to “negotiate” trust. Even if the two parties cannot finish the n-round information sharing process, they still could share some valuable information with each other during the process. In this way, information sharing is more efficient and cost-effective. •
Web services based implementation. Such implementation not only enables our scheme to be seamlessly integrated into the FEA reference model, but also enables our scheme to be easily integrated into a variety of existing E-Government applications.
5. INFORMAITON SHARING FRAMEWORK
Our information sharing framework has two parts: (1) an interest-based trust model; and (2) an information sharing protocol built on top of the trust model.
5.1 Assumptions
•
Centralized trust. We assume that all the agencies that want to share information with each other trust a single certification authority (CA) and any credentials issued by the CA. We assume the CA may delegate the authority to certify an attribute of a government agent to an agency. In this way, one agency may trust the credentials issued by another agency in an indirect fashion through a specific delegation credential issued by the CA. Two agencies need not to know each other in advance in order to be able to share information. •
Secure communication channels. We assume that all the agencies are using a secure communication channel to share information with each other since the shared information can be very sensitive. We assume the secure channel can ensure the confidentiality and integrity of the messages being transmitted. Denial-of-service attacks may happen but are out of the scope of this paper.
•
Intra-agency information sharing is in place. In fact, intra-agency information sharing can be easily achieved using the set of existing information sharing technologies overviewed in Section 4, where (a) centralized trust is naturally built; (b) when the agency’s information systems are centralized, access
9
control can be directly used to share information among units within the agency where agents playing a more responsible role can access more sensitive information than other agents; (c) when the agency’s information systems are distributed and of large scale, digital credentials and delegation can be used, where the amount of information that an agent can access is determined by “which attributes (e.g., a role) are associated with the agent?” Note that intra-agency information sharing typically does not need an interest-based trust model and existing trust models are usually good enough to enable effective intra-agency information sharing.
5.2 Design Goals and Requirements
The ultimate goal of our information sharing scheme is to transform the situation where “nobody wants to share information” to the situation where “everybody wants to proactively share information”, so that mutual trust can be rebuilt among agencies, differences between agencies can be tolerated, misunderstandings among agencies can be minimized, conflicts can be resolved, and effective information sharing can be achieved.
To achieve this goal, the information sharing scheme needs to satisfy the following requirements:
a) The trust building procedure, part of the information
sharing protocol, should provide enough incentives for agencies to do win-win information sharing and should discourage win-lose information sharing. In this way, more agencies will do win-win information sharing and “bad” agencies will be punished by doing win-lose information sharing.
b) To satisfy requirement (a), the information sharing
protocol must ensure that no agency can get significant amount of advantages during any steps of the protocol if he/she hides information or does not play honestly, because otherwise win-lose information sharing will be encouraged.
c) Requirement (b) indicates that during any step of the
information sharing protocol no significant amount of information can be disclosed, because otherwise the agency who receives the significant amount of information can abort and get significant amount of advantages.
d) Since no significant amount of information can be
disclosed in any step of the protocol, gradual information disclosing is required.
e) Requirements (b) and (d) indicate the disclosing of
information from agency A to agency B and from B to A should be interleaved with each other, because otherwise the agency who discloses the information first will be the loser even if he/she discloses the information gradually.
f) Requirements (a) and (e) indicate that trust building
and information exchange should be interleaved with each other, since validity and utility of information are part of trust, and trust is a precondition for information disclosure.
g) Fairness: The protocol should ensure that every
information-sharing procedure is fair, that is, no matter where the procedure stops (note that an agency may abort at any point of the procedure), the information or the interests gained by the two agencies (involved in the procedure) should be almost the same. From the perspective of trust building, fairness can be interpreted as: either the information is exchanged and the trust levels are increased, or the information is not exchanged at all and the trust levels are dropped, but nothing in between could happen. This requirement can be viewed as being deduced from requirement (a).
5.3 Information Sharing Policies
We use four types of information sharing policies in the information sharing protocol. They are information release (IR) policies, credential release (CR) policies, trust level adjustment (TLA) policies, and access control (AC) policies.
To illustrate the four types of policies, we need to first introduce a set of notations. In our protocol, we assume there are two agencies: A and B. We assume Ca1, …, Cam are the m credentials held by agency A; and Cb1, …, Cbn are the n credentials held by agency B. We assume Ua1, …, Uau are the u units of information owned by A; and Ub1, …, Ubv are the v units of information owned by B. We assume TL(A) is the trust level of A in the eyes of agency B, which indicates the degree to which B trusts A; and TL(B) is the trust level of B maintained by A. We assume the trust levels are quantified using an integer from 1 to 10, the simplest way. Although trust levels can be quantified in a more complicated way, numerical trust levels are enough to show the idea of our information sharing protocol. Finally, we assume utitlity(Uai) measures the utility level of information unit Uai to agency A, which indicates the degree to which Uai is useful to A. We assume utilities are quantified using an integer from 1 to 7.
Now we define the four types of information sharing policies one by one following an order that can best illustrate the relationships among these policies.
An information release policy is used to help agency A to determine when a specific information unit can be disclosed to agency B. An IR policy is simply a set of component policies. An example IR component policy is shown in Figure 4, where we can see that an IR component policy is composed of two parts: the condition part is a conjunction or disjunction of a set of predicates, and the action part specifies the information unit that can be
10
released when the condition is satisfied?” The condition of an IR component policy can consist of 0 to 4 predicates. When there are 0 predicates, there is no restriction to disclose the information unit. When there are 4 predicates, as Figure 4 shows, when agency A wants to determine whether to disclose an information unit to agency B, the first predicate is to check if B is “authorized” to access the information unit. How such checking should be done is specified by an access control policy which we will define shortly.
Condition: access-control-passed({Cb1, …, Cbp}, Uaj) AND received-and-valid(Ub1, …, Ubq) AND utility(Ub1, …, Ubq) >3 AND TL(B) >4 Action: release information unit Uaj to B
FIG. 4. An Example IR Component Policy
Condition: received-and-verified(Cb1, …, Cbp) AND attributes-carried({Cb1, …, Cbp}, {a1, …, ar}) AND received-and-valid(Ub1, …, Ubq) AND utility(Ub1, …, Ubq) >4
Action: increase the value of TL(B) by 1 FIG. 5. An Example TLA Component Policy Condition: received-and-valid(Ub1, …, Ubq) AND
utility(Ub1, …, Ubq) >3 AND TL(B) >5 AND
received-and-verified(Cb1, …, Cbp) AND attributes- carried({Cb1, …, Cbp}, {a1, …, ar})
Action: release credential Caj to B
FIG. 6. An Example CR Component Policy Condition: received-and-verified(Cb1, …, Cbp) AND
attributes-carried({Cb1, …, Cbp}, {a1, …, ar})
Action: B can access information unit Uaj
FIG. 7. An Example AC Component Policy
The second predicate is to verify the validity of the information units that agency A has already received from B through the (same) information sharing procedure so far. The third predicate is to check the utilities already earned by A through the protocol run. We assume the validity is checked and the utilities are measured by a human being. How to have a software agent check the validation and measure the utilities is one of our future
research topics. The fourth predicate is to check if A has built a certain level of trust in B.
In summary, the first predicate builds the lower level trust needed for cross-agency information sharing. The second and third predicates are to satisfy design requirements (a), (b), (c), (d), (e), and (g). The fourth predicate is to satisfy design requirement (f).
In our scheme, trust levels are dynamically maintained based on a specific trust level adjustment policy. An example TLA component policy is shown in Figure 5, where (a) the first and second predicates build the lower level trust on top of which interest-based trust can be built. The attribute-carried() predicate will help agency A build a level of certified-attribute-based trust in B. Note that this level of trust is exactly the type of trust built by existing trust management techniques (see Section 4). (b) The third and fourth predicates build an additional level of interest-based trust. Finally, it should be noticed that Figure 4 and Figure 5 show clearly how trust building and information exchange are interleaved with each other. In our scheme, the disclosure of credentials is also controlled by a credential release policy, since credentials may contain very sensitive attributes (e.g., the responsibility of a special FBI agent). An example CR component policy is shown in Figure 6, where we can see that whether to release a credential to B is not only dependent on whether B has released some sensitive credentials to A, but also dependent upon the level of interest-based trust A has in B, since releasing sensitive credentials may also hurt the interests of an agency. It should be noticed that our CR policies are different from existing trust negotiation policies [46].
Finally, an access control policy determines whether an agency is eligible or authorized to access an information unit owned by another agency. An example AC component policy is shown in Figure 7, where we can that our access control policies are almost the same as existing trust-based information access policies [44]. Nevertheless, it should be noticed that Figures 6 and 7 show that information exchange and credential negotiation are actually interleaved with each other. Figure 4, Figure 5, and Figure 6 show that trust building and credential negotiation are actually also interleaved with each other. 5.4 Information Sharing Protocol
The four types of information sharing policies clearly indicate how an information sharing protocol should run. In this section, instead of giving a formal specification of the protocol, we use an example to illustrate how the protocol works.
The example is shown in Figure 8, where the protocol run involves seven messages and six major procedures, namely P1, …, P6, after some of the messages. We assume
11
agency A initiates the information sharing procedure by asking B to disclose information unit Ub3. We assume A also discloses a credential, namely Ca1, together with the request. B processes this request via P1 as follows: (a) P1 will first use Ca1 to adjust the value of TL(A) according to a TLA component policy, we assume as a result, TL(A) is increased from 1 to 2. Next, P1 will check the set of IR component policies regarding Ub3. We assume {Ca1, Ca2, Ca3} are needed to access Ub3. So each access-control-passed() predicate in these component policies will be evaluated as FLASE. Next, to enable agency A to disclose Ca2 and Ca3 to B (so that B may disclose Ub3 to A), B needs also to disclose some information units and credentials to A so that the set of security requirements listed in Section 5.2 can be satisfied. Hence, B will disclose Ub1 and Cb1 in message 2. Here Ub1 can be disclosed because we assume there is an IR component policy regarding Ub1 allows B to do so based on Ca1 and TL(A). Moreover, B asks A to disclose Ua3.
When A receives message 2, P2 will first use Ub1 and Cb1 to adjust the value of TL(B). Next, P2 will check his/her IR and CR component policies to figure out which credentials and/or information units can be disclosed. We assume {Cb1, Cb2, Cb3} are needed to disclose Ua3, and Ua1 and Ca2 can be disclosed due to Ub1, Cb1, and the fact that TL(B) is increased.
When B receives message 3, P3 will do almost the same type of things as P2 does and as a result, Ub2 and Cb2 are disclosed. When A receives message 4, P4 will do almost the same type of things as P2 does and as a result, Ua2 and Ca3 are disclosed. When B receives message 5, P5 will do almost the same type of things as P3 does and as a result, Ub3 and Cb3 are disclosed. Finally, when A receives message 6, A will disclose Ua3 in message 7. Agency AAgency B[1] “I want Ub3”, Ca1[2] “I can only release U“I want Ub1”, Ub1, P1a3”, Cb1P2[3] “I can only release Ua1”, Ua1, “I can release Ca2now”, Ca2[4] “I can release Ub2”, Ub2, “I can release CP3b2now”, Cb2P4[5] “I can release Unow”, Ca2”, Ua2, “I can release Ca3a3[6] “I can release Ub3”, Ub3, “I can release CP5b3now”, Cb3P6[7] “I can release Ua3”, Ua3, FIG. 8. An Example Run of the Information Sharing Protocol
Although we will not formally specify our information sharing protocol, a formal specification of the protocol should be easily figured out based on the above example protocol run. 5.5 Discussion
First, some people may wonder “how can agency A know after receiving message 1 that Ua1 and Ca2 will be needed to enable B to finally disclose Ub3?” The answer has two parts: (a) agency B can tell A via message 1 which kind of credentials is needed for B to disclose Ub3. B can figure out this information based on his/her IR and CR component policies using the techniques developed in [46]. (b) Ua1 is chosen primarily because of the relation between Ua1 and Ua3. Since B wants Ua3, so information related to Ua3 should be able to increase the interests of B and encourage B to increase his/her trust in A. For example, Ua1 can be an approximation (or estimation) of Ua3.
Second, our information sharing scheme has three key features: (1) it uses an interest-based trust model where utilities of shared information units are part of the mutual trust. (2) It interleaves trust negotiation and information exchange, where not only trust negotiation is gradually done but also information sharing is gradually done. (3) It satisfies the set of security requirements identified in Section 5.2. Therefore, our information sharing scheme can meet the design goals identified in Section 5.2.
6. PROTOTYPE IMPLEMENTATION
To be completely compatible with the FEA, we implement our information sharing scheme using XML web services. Web services are an emerging, state-of-the-art technology that provides the perfect technology for integrating and sharing information among agencies. One of its primary features is interoperability of two systems. Web services are a critical part of both the FEA conceptual process reference model and the interoperability reference model, since web services can enable government agencies with different types of information systems (e.g., different data management schemes, different operating systems, different network protocols) to communicate and interact with each other in a platform transparent fashion.
6.1. Components of Web Services
The components of XML web services include applications, UDDI, WSDL, and SOAP [16]. Applications may be clients of XML web services provided by another
12
application, providers of XML web services to other applications, or both [31]. UDDI, Universal Description Discovery and Integration, provides a registry of available XML web services. The Web Service Description Language (WSDL) provides an XML based description of XML web services and how to interact with them [14]. The WSDL is the interface definition language of XML web services. Finally, SOAP, or Simple Object Access Protocol, is a lightweight remote-procedure-call protocol that uses XML to format messages and HTTP to transmit messages [3].
The general procedure of using a web service starts when a client searches for the web service from an UDDI registry. The UDDI registry in some sense is equivalent to the yellow pages in a phone book. However, the UDDI registry is not a completely secure registry and in fact is a public registry. Many government officials are very reluctant to populate a public registry of web services using the UDDI standard [29]. As a result, government agencies that wish to share their services with each other will share the service description information through a secure communication channel. In either way, the client will be able to obtain a description of the web service, which is written in WSDL, through a private or public registry. Then, the client can retrieve the description (written in WSDL), parse the XML data contained in the description, and learn the URL of the web service and the right ways to call the web service using SOAP. At the web server end, each SOAP request will be first received by the SOAP listener then forwarded to a specific request processor, called the WSDL server, which will parse the XML data contained in the SOAP request. A set of service requests which can be understood by the server application will be generated after the XML data are parsed, and the set of service requests will be sent to the server application for processing.
6.2. Microsoft .NET XML Web Services
We implement our information sharing scheme using Microsoft .NET, which provides a very powerful, mobile, and developer friendly infrastructure for creating a web service. Many real world government agencies have already deployed some .NET web services.
Some important components of XML web services in .NET include the .ASMX file extension, the .ASPX file extension, and the WSDL. Initially, the web service itself is implemented using the .ASMX file extension. Essentially, .ASMX files are based on ASP .NET. In these .ASMX files, besides the code that implements each web method of the web service, the set of web methods that compose the web service are also defined using a set of After the (web service) provider creates all of its functions or web methods in the .ASMX files, it creates a WSDL description of the web service, where the client can find the exposed methods of the web service, the parameters and return types that these methods expose, and any other exposed information. The WSDL description is generated based on the set of [SoapHeader(\"Credentials\[WebMethod] true)]public return_type function_name() { //Code used to implement web method } FIG. 9. SOAP Header Format in .NET 6.4. How to Use XML Web Services to Implement Our Information Sharing Protocol? In general, we can use XML web services to implement our information sharing protocol as follows: (a) each agency X provides a set of information-sharing web services to the agencies who may want to share information with agency X. The WSDL descriptions of these web services should be composed and disclosed to an agency that wants to share information with X in a secure way, since such descriptions may themselves contain sensitive information and should only be readable to certain authorized agencies. (b) In addition, each agency X provides a specific information- sharing-support web service to her own agents. Since this service (i.e., the URL) is pre-known to every agent of X, no WSDL description or registration is needed. (c) When an agent of agency X wants to get a specific information unit U from agency Y, the agent will first authenticate himself/herself to the information-sharing-support service (“support service” for short) provided by X. Next, the agent will use a standardized interface (provided by the support service) to ask the support service to initiate an information sharing procedure with agency Y. However, to have an error-free information sharing procedure with Y, the support service must first understand which kinds of information sharing services are provided by Y and how these services should be called (or used) in terms of the procedures and the message format that need to be followed. Hence, before the support service starts the information sharing procedure, it will first search for the WSDL descriptions of the services provided by Y. Next, the support service will compose a request message 13 needs to be involved in the information sharing procedure, the service of Y will notify the agent and prepare an interface for the agent to jump in. (e) In the following, the agent of Y and the agent of X can follow the information sharing protocol (e.g., the procedure shown in Figure 8) in a naïve manner, except that in the implementation the support service functions as the “proxy” for the agent of X, and the information sharing service functions as the “proxy” for the agent of Y. 6.5. An Example Implementation To better understand how exactly web services can be used for two agencies to share information (with each other) within the federal enterprise architecture framework, we implemented a simple information sharing system prototype using XML web services for a real world information sharing application between two agencies. In particular, we use this prototype to support FBI and CIA to share anti-terrorism information with each other1. However, for simplicity, we implement the FBI’s information sharing web service, but do not implement the support service of the CIA; instead, we make a CIA agent directly call the FBI’s service. (Note that this implementation can be easily extended to include all the components we mentioned in Section 6.4.) In this prototype, we assume both the FBI and the CIA contain pertinent information with regards to anti-terrorism. For example, the FBI may maintain such anti-terrorism information as criminal lists and the CIA may maintain such anti-terrorism information as a set of intelligence gatherings1. The steps that the two agencies will take in sharing information with each other are depicted in Figure 10. In particular, we assume a CIA agent wants to get an information unit from the FBI. So initially, the FBI will create an information-sharing web service and specify the web methods for the service. Then the FBI will send the WSDL description of the service (or the set of web methods) to a secure web site (or a secure channel) for the CIA agent to retrieve. A sample of the WSDL service description is shown in Figure 11. Moreover, one way to create the secure channel in the federal enterprise architecture is to use a trusted broker. In 1 The FBI and CIA used in this implementation are used only as example agencies. The real government agencies may contain different names, perform different responsibilities, and contain different types of information. particular, when one agency X needs to share information with another agency Y, it can simply send its WSDL service description along with the agency it wants to share information with, to a broker. The broker will allow an agency to access the service description only if the agency is authenticated as Y. 1. FBI, the web service provider agency develops its information sharing policies and establishes the corresponding web methods 2. The FBI passes the WSDL description of its information sharing web service to a secure channel 3. The CIA agent, i.e., the web service client,retrieves the WSDL description 4. The client parses the WSDL description, calls the corresponding web methods of the information sharing web service to initiate and interact with the FBI’s service in such a way that the information sharing protocol can be successful executed FIG. 10. Steps of Information Sharing between Two Agencies Using Web Services FIG. 11. WSDL Description of an Information-Sharing Web Service Once the client (on behalf of the CIA agent) retrieves the WSDL description, using an XML parser, it will analyze the service description and determine the input arguments for the web methods that it needs to invoke during the protocol run via SOAP messaging. 14 6.5. Key Features The prototype implementation has several interesting features, which are as follows. These features indicate the benefits of using XML web services. • It supports two-way information sharing. When the CIA client calls a FBI web method, the client can disclose some credentials or information to the FBI. On the other hand, the execution of a FBI web method can disclose some credentials or information to the CIA agent in several ways. For example, the FBI web method can show an information unit or a credential directly on the interface (i.e., a web browser) used by the CIA agent. For another instance, the FBI web method can guide the CIA agent to download an information unit or a credential via the web browser. • It provides multiple web methods for building trust. • It exploits SOAP header authentication to counter session hijack attacks. The idea of SOAP header authentication is shown in Figure 9. That is, a web method can be executed only if the credentials (e.g., a password) provided by the client can authenticate the client. Using this technique, the FBI web service can enforce dynamic authentication for each web method in such a way that the attacker will not succeed in replaying web methods or hijacking an ongoing information sharing session. • It provides privacy for using web methods. That is, clients not qualified to run a web method (due to trust and authorizations) will never be able to see “what is the web method about?” or “how should the web method be used?” • It provides a design for peer-to-peer cross-agency information sharing. • It provides a powerful, easy-to-use interface with .NET technology. • It integrates information sharing and data management. Both the CIA client and the FBI web service manage the information units they may share in a MS ACCESS database. 6.6. Limitations Nevertheless, at this stage, the prototype is still preliminary and it has several limitations, which are as follows. The last three limitations can be overcome by better engineering, while the first limitation needs future research, which we will address in Section 7. 1) It requires human interaction. 2) It is only an emulation of the real world information sharing activities between FBI and CIA. It uses emulated data instead of real data. 3) It uses a small, centralized database. 4) Using a broker to create a secure channel for WSDL service descriptions distribution introduces some extra overhead compared with using UDDI. 7. CONCLUSION AND FUTURE RESEARCH Although trust-based information access is well studied in the literature, the existing trust models, which are based on certified attributes, cannot support effective information sharing among government agencies, which requires an interest-based trust model. To solve this information sharing problem, this paper proposes an innovative interest-based trust model and a novel information sharing protocol, where a family of information sharing policies are integrated, and information exchange and trust negotiation are interleaved with and interdependent upon each other. In addition, an implementation of this protocol is presented using the emerging technology of XML Web Services. The implementation is totally compatible with the Federal Enterprise Architecture reference models and can be directly integrated into existing E-Government systems. We believe our cross-agency information sharing scheme can transform the situation where “nobody wants to share information” to the situation where “everybody wants to proactively share information”, so that the mutual trust can be rebuilt among agencies, differences between agencies can be tolerated, misunderstandings among agencies can be minimized, conflicts can be resolved, and effective information sharing can be achieved. With secure, effective information sharing, the agencies’ ability to predict attacks and preempt them can be significantly enhanced. In addition, the following future research issues are identified and successful resolution of these research issues can further improve the agencies’ ability to effectively and efficiently share information with each other. 7.1 Develop Privacy Preserving UDDI Registry One way to improve upon the existing implementation is to develop a privacy preserving UDDI registry. With such a UDDI, an agency A will be able to register his/her information sharing services without losing privacy, in the sense that his/her services will only be visible to the agencies that B would like to share information with. In this way, agencies can enjoy the simplicity and efficiency of UDDI without losing confidentiality; and no dedicated secure service notice facilities are needed. 7.2. Develop Automated Software Agents for Trust Negotiation Another way to improve upon the existing implementation is to develop automated or semi-automated software agents who negotiate trust and share sensitive 15 information with each other on behalf of the corresponding agencies. In this way, human being can be completely or partially relieved from the labors (or efforts) needed to run the information sharing protocol; and more efficient and timely information sharing may be achieved. More timely information sharing can mean more timely detection of terrorism attacks and less damage caused by such attacks. A key research issue in using software agents is how to enable a software agent to intelligently evaluate the validity and utility of a piece of information. Although this issue is out of the scope of this paper, we believe artificial intelligence and natural language processing would play an important role in this research. 7.3. Cookie Management for Incremental Trust Negotiation Another interesting future research area would be to develop cookie management between two agencies. Cookies are information stored on your own computer for future use after some www sessions. With the development of cookie management, two agencies that have already built certain level of trust with one another don’t have to redevelop the mutual trust. Instead, they could use cookies to remember the current trust state for future reuse. In this way, information sharing can be made quicker and more efficient. Nevertheless, a drawback of using cookies is that the security risk is increased. 7.4. Direct Querying According to Previously Stored Information In relation to this research, another efficient way for each agency to share information with one another is for them to directly “query” the other agencies’ databases according to the database scheme information they were already given. For instance, if two agencies have already shared information with one another, then each agency would have already obtained substantial scheme information about the other agency’s databases. Such scheme information may enable both agencies to compose accurate, executable queries. By sending a query directly from one agency to another, an agency will be able to pinpoint the information he/she wants to get in a finer granularity. REFERENCES 1 Agrawal R., Evfimievski A., Srikant, R. (2003). Information Sharing Across Private Databases. In Proc. ACM SIGMOD 2003, San Diego, CA. 2 Ahn, G., Chu, B., Zhang, L. (2001). A Role-Based Framework for Healthcare Information Systems. In Proc. ACM Symposium on Access Control Models and Technologies, pages 153-162. 3 Ajmani, S., Morris R., Liskov, B. (2001). A Trusted Third-Party Computation Service. Technical report MIT-LCS-TR-847, MIT 4 Akella, Y., Bilello, M., Dawson, S., De Capitani di Vimercati, S., Samarati, P., Lincoln, P., Tan, Y., Wiederhold, G. (2000). Secure Access Wrapper: Mediating Security between Heterogeneous Databases. In Proc. of the DARPA Information Survivability Conference and Exposition, Hilton Head, South Carolina. 5 Asokan, N., Shoup, V. (1998). Asynchronous Protocols for Optimistic Fair Exchange. Proc. IEEE Symposium on Research in Security and Privacy, pages 86-99, CA. 6 Augustyniak, M. (2002). SAMS Teach Yourself .NET XML Web Services in 24 Hours. SAMS Publishing, Indianapolis, Indiana. 7 April, C. A., Fonseca, B. (2002). Monumental Mission. InfoWorld, http://www.infoworld.com/article/02/09/06/020909ctgovt_1.html 8 Bao, F., Deng, R.H., Mao, W. (1998). Efficient and Practical Fair Exchange Protocols with Off-Line TTP. Proc. IEEE Symposium on Research in Security and Privacy, pages 86-99, CA. 9 Ben-Or, M., Goldreich, O., Micali, S., Rivest, R.L. (1990). A Fair Protocol for Signing Contracts. IEEE Transactions on Information Theory, Vol. 36, No. 1, pp 40-46. 10 Birbeck, M., Diamond, J., Duckett, J., Gudmundsson, O.G., Kobak, P., Lenz, E., Livingstone, S., Marcus, D., Mohr, S., Ozu, N., Pinnock, J., Visco, K., Watt, A., Williams, K., Zaev, Z., (2001). Professional XML, 2nd Ed., Wrox Press, Birmingham, UK. 11 Blaze, M., Feigenbaum, J., Ioannidis, J., Keromytis, A.D. (1996). Decentralized Trust Management. In Proc. the 1996 IEEE Symposium on Security and Privacy, pages 164-173. 12 Blaze, M., Feigenbaum, J., Ioannidis, J., Keromytis, A.D. (1999). The Keynote Trust-Management System, version 2, IETF RFC 2704. 13 Bush, G. W. (2002). Presidential Memo on the Importance of E-Government: Memorandum for the Heads of Executive Departments and Agencies. Office of the Press Secretary. http://www.whitehouse.gov/news/releases/2002/07/20020710-6.html. 14 Cameron, D. (1998). E-Commerce Security Strategies: Protecting the Enterprise. Computer Technology Research Group, Charleston, South Carolina. 15 Carbone, G., Stoddard, G. (2001). Open Source Enterprise Solutions. Wiley Computer Publishing, New York, New York. 16 Champion, M. E. (2002). Web Services Architecture: W3C Working Draft. W3C. http://www.w3.org/TR/2002/WD-ws-arch-20021114. 17 Chu, Y., Feigenbaum, J., LaMacchia, B., Resnick, P., Strauss, M. (1997). REFEREE: Trust Management for Web Applications. World Wide Web Journal, Vol. 2, pages 706-734. 18 Clarke, D., Elien, J., Ellison, C., Fredette, M., Morcos, A., Rivest, R.L. (2001). Certificate Chain Discovery in SPKI/SDSI. Journal of Computer Security, Vol. 9, No. 4, pp 285-322. 19 Connel, P. O., Gray, E., Jensen, C., Seigneur, J.-M., Weber, S., Yong, C. (2002). Towards a Framework for Assessing Trust-Based Admission Control in Collaborative Ad Hoc 16Applications. Informatics and Mathematical Modeling, Technical University of Denmark, Lyngby, Denmark. 20 Cox, B., Sirbu, M., Tygar, D., (1995). NetBill Security and Transaction Protocol, Proc. the First USENIX Workshop on Electronic Commerce, pages 77-88, New York. 21 Coyne, E.J., Feinstein, H.L., Sandhu, R., Youman, C.E. (1996). Role-Based Access Control Models. IEEE Computer, Vol. 29, No. 2. 22 Cummins, F. A., (2002). Enterprise Integration: An Architecture for Enterprise Application Systems Integration. Wiley Computer Publishing, New York, New York. 23 Damiani, E., Paraboschi, S., Vimercati, S. (2002). A Reputation-Based Approach for Choosing Reliable Resources in Peer-to-Peer Networks. Proc. ACM Conference on Computer and Communications Security, Washington, DC, USA, pp207-216. 24 Deng, R.H., Gong, L., Lazer, A.A., Wang, W. (1996). Practical Protocols for Certified Electronic Mail. Journal of Network and System Management, Vol. 4, No. 3. 25 Digital Signature Initiative. (1997). The Digital Signature Trust Management Architecture. W3C, http://www.w3.org/Dsig/DsigTrustSystem.html. 26 R Dodds, L. (2002). In a Lather About Security. O’Reilly Network. http://ww.xml.com/pub/a/2002/02/27/security-lather.html. 27 Donovan, J. (2002). U.S.: Congress Probes FBI And CIA Intelligence Failures. RadioFreeEurope. http://ww.rferl.org/nca/features/2002/06/05062002152908.asp. 28 Ettinger, J.E. Ed. (1993). Information Security. Chapman & Hall, London, United Kingdom. 29 Ewald, T. (2002). The Web Services Idea. Microsoft Corporation. http://msdn.microsoft.com/webservices/understanding/readme/default.aspx. 30 Farah, J. (2002). U.S. Government Doesn’t Trust Americans. WorldNetDaily. http://www.worldnetdaily.com/news/printer-friendly.asp?ARTICLE_ID=28308. 31 Federal CIO Council. (2002). E-Gov Enterprise Architecture Guidance(Common Reference Model. FEA Working Group. http://ww.cio.gov/documents/ E-Gov_Guidance_July_25_Final_Draft_2_0a.pdf. 32 Federal Enterprise Architecture Program Management Office. (2002). The Business Reference Model Version 1.0:A Foundation for Government Wide Improvement. Federal Architecture Program Management Office. http://www.feapmo.gov/resources/fea_brm_release_document_rev_1.pdf. 33 Federal Enterprise Architecture Program Management Office. (2002). The Business Reference Model Version 1.0:Agency Briefing Federal Architecture Program Management Office. http://www.feapmo.gov/fea_downloads.htm. 34 Federal Enterprise Architecture Program Management Office. (2002). Draft Business Reference Model Common Response Document. Federal Architecture Program Management Office.. http://www.feapmo.gov/fea_downloads.htm. 35 Federal Enterprise Architecture Program Management Office. (2002). Solution Architects Working Group: Overview of Missions and Objectives. Federal Architecture Program Management Office. http://xml.gov/presentations/sawg/overview.ppt. 36 Federal Enterprise Architecture Program Management Office. (2002). Twenty Four Presidential Priority E-Govt Initiatives. Federal Architecture Program Management Office. http://www.feapmo.gov/fea.htm. 37 Fedorov, A. (2002). A Programmer’s Guide to .NET. Addison-Wesley, London, United Kingdom. 38 Fernandez, E. B. (2002) Web Services Security: Current status and the future. Web Services Architect. http://www.webservicesarchitect.com/content/ articles/fernandez01.asp. 39 Franklin, M.K., Reiter, M.K. (1997). Fair Exchange with a Semi-Trusted Third Party. Proc. 4th ACM Conference on Computer and Communications Security, pages 1-7, Zurich, Switzerland. 40 French, M. (2003) CIA, FBI Wrangle Over Threat Center. Federal Computer Week. http://www.fcw.com/fcw/articles/2003/0421/web-ciafbi-04-23-03.asp. 41 Gasser, M., McDermott, E. (1990). An Architecture for Practical Delegation in a Distributed System. IEEE Symposium on Research in Security and Privacy, Oakland, CA. 42 Goldreich, O. (2001). Secure Multi-Party Computation. Working Draft, Version 1.3. 43 Gollmann, D., Zhou, J. (1996). A Fair Non-Repudiation Protocol. Proc. IEEE Symposium on Research in Security and Privacy, pages 55-61, Oakland, CA. 44 Grosof, B.N., Feigenbaum, J., Li, N. (2003). Delegation Logic: A Logic-Based Approach to Distributed Authorization. ACM Transactions on Information and Systems Security, Vol. 6, No. 1, pp 128-171. 45 Hailes, S., Rahman, A. (1997). A Distributed Trust Model. Proc. New Security Paradigm Workshop, ACM. 46 Hess, A., Jacobson, J., Jarvis, R., Seamons, K.E., Smith, B., Yu, L. Yu, T., Winslett, M. (2002). Negotiating Trust on the Web. IEEE Internet Computing. 47 Hoque, F. (2000). e-Enterprise: Business Models, Architecture, and Components. Cambridge University Press, New York, New York. 48 Jarrell, D. (2002) The Value of Information Sharing. Federal Computer Incident Response Center. http://www.fedcirc.gov. 49 Jefferies, N., Mitchell, C., Walker, M. (1995). A Proposed Architecture for Trusted Third Party Services. In Cryptography Policy and Algorithms Conference, Springer LNCS Volume 1029, pp 98-104. 50 Jones, V.E., Seamons, K.E., Winsborough, W.H. (1999). Automated Trust Negotiation. Open Group Conference, Montreal, Quebec. 51 Josang, A. (1999). An Algebra for Assessing Trust in Certification Chains. Proc. Network and Distributed Systems Security Symposium, The Internet Society. 52 Josang, A. (1997). The Right Type of Trust for Distributed Systems. Proc. New Security Paradigm Workshop, ACM. 1753 Kaler, C. Ed. (2002). Web Services Security: WS-Security. IBM. http://www-106.ibm.com/developerworks/webservices/library/ws-secure/. 54 Krill, Paul. (2002). UDDI Endures Reality Check Debate. InfoWorld. http://www.infoworld.com/article/02/04/26/020426hnuddi_1.html. 55 Liu, L., Xiong, L. (2003). A Reputation-Based Trust Model for Peer-to-Peer E-Commerce Communities. Proc. 4th ACM Conference on Electronic Commerce, San Diego. 56 Microsoft Visual Studio. NET. (2002). XML Web Services Security. Microsoft. http://msdn.microsoft.com/vstudio/techinfo/article/XMLwebservices/security.asp. 57 Naor, M., Nissim, K. (2001). Communication Preserving Protocols for Secure Function Evaluation. In Proc. of the ACM Symposium on Theory of Computing, pages 590-599. 58 Naor, M., Pinkas, B., Summer, R. (1999). Privacy Preserving Auctions and Mechanism Design. Proc. of the 1st ACM Conference on Electronic Commerce, pages 129-139, Denver, Colorado. 59 Network of Reform Groups. (1999). The Effective National Drug Strategy 1999. Network of Reform Groups. http://www.csdp.org/edcs/page45.htm. 60 OMB Watcher. (2002). NY State Confidential Memorandum. OMB Watcher. http://www.ombwatch.org/info/2001/NYSinventory.html. 61 Open Resource Group. (1999). Technical Reference Model in Detail. The Open Group. http://www.opengroup.org/architecture/togaf7/presents/togaf_ovu.pdf. 62 Opplinger, R. (1999). Security Technologies for the World Wide Web. Artech House, Boston, MA. 63 Osborn, S. L. (2002). Integrating Role Graphs: A Tool for Security Integration. Data and Knowledge Engineering, Vol. 43, No. 3, pp. 317-333. 64 Pipkin, D.L. (2000). Information Security: Protecting the Global Enterprise. Prentice Hall, Upper Saddle River, New Jersey. 65 Scheer, A.-W. (1998). ARIS – Business Process Frameworks. Springer, Berlin, Germany. 66 Seamons, K.E., Winslett, M., Yu, T. (2001). Limiting the Disclosure of Access Control Policies During Automated Trust Negotiation. Network and Distributed System Security Symposium. San Diego, CA. 67 Shirky, C. (2002) Web Services-An Executive Summary. O’Reilly Network. http://www.oreillynet.com/pub/a/webservices/2002/04/12/execreport.html. 68 Skonnard, A. (2002). Understanding XML Namespaces. DevelopMentor. http://msdn.microsoft.com/msdnmag/issues/01/07/xml/default.aspx. 69 Tabor, R. (2002). Microsoft .NET XML Web Services. SAMS Publishing, Indianapolis, Indiana. 70 Tygar, J.D. (1998). Atomicity Versus Anonymity: Distributed Transactions for Electronic Commerce, thProc. 24 VLDB Conference, pages 1-12, New York. 71 Ulanoff, L. (2002). An Insider’s Look at Homeland Security and Technology. PC Magazine. http://www.pcmag.com/article2/0,4149,652467,00.asp. 72 Washington Technology. (2002). Understanding the Federal Enterprise Architecture. Washington Technology. http://216.70.54.91/ad_sup/blueprintl. 73 Wolter, R. (2001). XML Web Services Basics. Microsoft Corporation. http://msdn.microsoft.com/library/default.asp?url=/library/en-us/dnwebsrv/html/webservbasics.asp. 74 Yao, A.C. (1986). How to Generate and Exchange Secrets. Proc. of the 27th Annual Symposium on Foundations of Computer Science, pages 162-167. Toronto, Canada. 75 York, S. (1997). Battling the Cyberthreat. Wabash Magazine. http://www.wabash.edu/magazine/1997/fall/features/face_of_conflict/cyber_threat.htm. 76 http://www.cia.gov. 77 http://www.fbi.gov 18 因篇幅问题不能全部显示，请点此查看更多更全内容