This paper presents a longitudinal study investigating how the General Data Protection Regulation (GDPR) compliance of website privacy policies has evolved over a fiveyear period. Using an automated privacy policy evaluation tool, we assessed ten core GDPR factors across a corpus of websites originally analyzed in 2020 and re-evaluated in 2025. Our analysis reveals a mixed progression: while user-facing compliance measures such as consent, data retention notification, and data sharing transparency showed measurable improvement, technically oriented factors-such as breach notification and data encryption-experienced a decline in explicit disclosure. These findings suggest a broader trend in which privacy policies increasingly emphasize legal rights and visible consent mechanisms, while de-emphasizing backend technical safeguards. The results point to a split in compliance communication, possibly influenced by regulatory clarity, enforcement pressure, and shifts in organizational privacy strategy. This study underscores the importance of continued policy auditing and the need for complementary methods that bridge the gap between stated policy and implemented practice in the context of evolving digital governance frameworks.

Online privacy policies are lengthy and hard to read, yet are profoundly important as they communicate the practices of an organization pertaining to user data privacy. Privacy Enhancing Technologies, or PETs, seek to inform users by summarizing these privacy policies. Efforts in the research and development of such PETs, however, have largely been limited to tools that recap the policy or visualize it. We present the next generation of our research and publicly available tool, PrivacyCheck v3, that utilizes machine learning to inform and empower users with respect to privacy policies. PrivacyCheck v3 adds capabilities that are commonly absent from similar PETs on the web. In particular, it adds the ability to (1) find the competitors of an organization with Alexa traffic analysis and compare policies across them, (2) follow privacy policies to which the user has agreed and notify the user when policies change, (3) track policies over time and report how often policies change and their trends, (4) automatically find privacy policies in domains, and (5) provide a bird's-eye view of privacy policies. The new features of PrivacyCheck not only inform users about details of privacy policies, but also empower them to understand privacy policies at a higher level, make informed decisions, and even select competitors with better privacy policies.

Studies have shown website privacy policies are too long and hard to comprehend for their target audience. These studies and a more recent body of research that utilizes machine learning and natural language processing to automatically summarize privacy policies greatly benefit, if not rely on, corpora of privacy policies collected from the web. While there have been smaller annotated corpora of web privacy policies made public, we are not aware of any large publicly available corpus. We use DMOZ, a massive open-content directory of the web, and its manually categorized 1.5 million websites, to collect hundreds of thousands of privacy policies associated with their categories, enabling research on privacy policies across different categories/market sectors. We review the statistics of this corpus and make it available for research. We also obtain valuable insights about privacy policies, e.g., which websites post them less often. Our corpus of web privacy policies is a valuable tool at the researchers' disposal to investigate privacy policies. For example, it facilitates comparison among different methods of privacy policy summarization by providing a benchmark, and can be used in unsupervised machine learning to summarize privacy policies.

The presence of disinformation and fake news on the Internet and especially social media has become a major concern. Prime examples of such fake news surged in the 2016 U.S. presidential election cycle and the COVID-19 pandemic. We quantify sentiment differences between true and fake news on social media using a diverse body of datasets from the literature that contains about 100K previously labeled true and fake news. We also experiment with a variety of sentiment analysis tools. We model the association between sentiment and veracity as conditional probability and also leverage statistical hypothesis testing to uncover the relationship between sentiment and veracity. With a significance level of 99.999%, we observe a statistically significant relationship between negative sentiment and fake news and between positive sentiment and true news. The degree of association, as measured by Goodman and Kruskal's gamma, ranges between. 037 to. 475. Finally, we make our data and code publicly available to support reproducibility. Our results assist in the development of automatic fake news detectors.

Despite the efforts to regulate privacy policies to protect user privacy, these policies remain lengthy and hard to comprehend. Powered by machine learning, our publicly available browser extension, PrivacyCheck v2, automatically summarizes any privacy policy by answering 20 questions based upon User Control and the General Data Protection Regulation. Furthermore, PrivacyCheck v2 incorporates a competitor analysis tool that highlights the top competitors with the best privacy policies in the same market sector. PrivacyCheck v2 enhances the users' understanding of privacy policies and empowers them to make informed decisions when it comes to selecting services with better privacy policies.

In this paper, we leverage previous work in the Identity Ecosystem, a Bayesian network mathematical representation of a person's identity, to create a framework to evaluate identity protection systems. Information dynamic is considered and a protection game is formed given that the owner and the attacker both gain some level of control over the status of other PII within the dynamic Identity Ecosystem. We present a policy iteration algorithm to solve the optimal policy for the game and discuss its convergence. Finally, an evaluation and comparison of identity protection strategies is provided given that an optimal policy is used against different protection policies. This study is aimed to understand the evolutionary process of identity theft and provide a framework for evaluating different identity protection strategies and future privacy protection system.

Personally Identifiable Information (PII) is often used to perform authentication and acts as a gateway to personal and organizational information. One weak link in the architecture of identity management services is sufficient to cause exposure and risk identity. Recently, we have witnessed a shift in identity management solutions with the growth of blockchain. Blockchain—the decentralized ledger system—provides a unique answer addressing security and privacy with its embedded immutability. In a blockchain-based identity solution, the user is given the control of his/her identity by storing personal information on his/her device and having the choice of identity verification document used later to create blockchain attestations. Yet, the blockchain technology alone is not enough to produce a better identity solution. The user cannot make informed decisions as to which identity verification document to choose if he/she is not presented with tangible guidelines. In the absence of scientifically created practical guidelines, these solutions and the choices they offer may become overwhelming and even defeat the purpose of providing a more secure identity solution.

Social media are providing a huge amount of information, in scales never possible before. Sentiment analysis is a powerful tool that uses social media information to predict various target domains (e.g., the stock market). However, social media information may or may not come from trustworthy users. To utilize this information, a very first critical problem to solve is to filter credible and trustworthy information from contaminated data, advertisements, or scams. We investigate different aspects of a social media user to score his/her trustworthiness and credibility. Furthermore, we provide suggestions on how to improve trustworthiness on social media by analyzing the contribution of each trust score. We apply trust scores to filter the tweets related to the stock market as an example target domain. While social media sentiment analysis has been on the rise over the past decade, our trust filters enhance conventional sentiment analysis methods and provide more accurate prediction of the target domain, here, the stock market. We argue that while it is a failing to ignore the information social media provide, effectively trusting nobody, it is an equal failing to trust everybody on social media too: Our filters seek to identify whom to trust.

Prior research shows that only a tiny percentage of users actually read the online privacy policies they implicitly agree to while using a website. Prior research also suggests that users ignore privacy policies because these policies are lengthy and, on average, require 2 years of college education to comprehend. We propose a novel technique that tackles this problem by automatically extracting summaries of online privacy policies. We use data mining models to analyze the text of privacy policies and answer 10 basic questions concerning the privacy and security of user data, what information is gathered from them, and how this information is used. In order to train the data mining models, we thoroughly study privacy policies of 400 companies (considering 10% of all listings on NYSE, Nasdaq, and AMEX stock markets) across industries. Our free Chrome browser extension, PrivacyCheck, utilizes the data mining models to summarize any HTML page that contains a privacy policy. PrivacyCheck stands out from currently available counterparts because it is readily applicable on any online privacy policy. Cross-validation results show that PrivacyCheck summaries are accurate 40% to 73% of the time. Over 400 independent Chrome users are currently using PrivacyCheck.

Victim-enabled identity theft is a crime in which an individual victim is deceived into providing their personally identifying information (PII) to a criminal to facilitate its theft and/or misuse. In this paper we analyse a particular victim- enabled tax-related identity theft scheme recently reported in Australia, which has also been reported, in a slightly different guise, in the US. We find that this scheme, and others like it, are best understood when studied from both the perpetrator's and the victim's points of view. The criminal perspective and business practices have been captured and analysed in the Identity Threat Assessment and Prediction (ITAP) model developed by the Center for Identity at The University of Texas (UT CID). The victim perspective has been captured from multiple victim case files captured by IDCARE. The research findings support the view that combining perspectives enhances the analytical value of a threat assessment and prediction model. The multi-actor nature of victim-enabled identity theft complements the methodological approach adopted in the paper, and provides new insights on a growing form of identity theft that can inform future prevention and detection response strategies.

Online transactions require a fundamental relationship between users and resource providers (e.g., retailers, banks, social media networks) built on trust; both users and providers must believe the person or organization they are interacting with is who they say they are. Yet with each passing year, major data breaches and other identity-related cybercrimes become a daily way of life, and existing methods of user identity authentication are lacking. Furthermore, much research on identity trustworthiness focuses on the user's perspective, whereas resource providers receive less attention. Therefore, the current research investigated how providers can increase the likelihood their users' identities are trustworthy. Leveraging concepts from existing research, the user-provider trust relationship is modeled with different transaction contexts and attributes of identity. The model was analyzed for two aspects of user-identity trustworthiness - reliability and authenticity - with a significant set of actual user identities obtained from the U.S. Department of Homeland Security. Overall, this research finds that resource providers can significantly increase confidence in user-identity trustworthiness by simply collecting a limited amount of user-identity attributes.

No abstract available.

Individuals declare their identities to online network providers with credentials such as usernames, passwords, and email addresses. To obtain these credentials from providers, users enroll by providing identity attributes, or collections of personal identifiable information (PII), such as phone numbers. Credentials vary in trustworthiness, and thus, so do identities. In search of better methods for increasing trustworthiness, we present a computational model of identity attributes described as an Identity Ecosystem to determine which are most vulnerable to malicious users. Using existing data from the U.S. Army and Department of Defense, wecmodel relationships between attributes as transition probabilities and analyze the long-run probability of all connected attributes being affected by one compromised attribute. This approach allows the provider to determine how best to weight relationships between attributes and thereby become more secure.

The concept of trust as presented here focuses on the trustworthiness, or reliability, of information and information sources. Decision makers, or agents, can create judgments based on previous experience with other agents and by reputation information received from allied agents. These judgments, or trust assessments, are used to predict the behavior of other agents and analyze the trustworthiness, truthfulness, or quality of information. Research concepts have been developed within the trust community, and they are most commonly applied to multi-agent systems research. This work attempts to show that trust research can be directly applied to security problems. Modern trust concepts enforce soft security which can be applied in addition to conventional security methods to build a more robust secure system. This article examines two trust based techniques and demonstrates their basic effectiveness using empirical experimentation. These techniques are then applied in a case study drawn from a more robust domain concerning confidential message transmission. The benefits of applying trust-based techniques to secure a system are measurable, and the costs associated with such techniques are scalable to even the most resource constrained systems.

When developing sophisticated multi-agent systems whose behaviors include collaboration, negotiation, and conflict resolution, analyzing and (empirically) verifying agent system behavior is a challenging task. To aid the developer in such tasks, this

The structure of a software architecture strongly influences the architecture's ability to prescribe systems satisfying functional requirements, non functional requirements, and overall qualities such as maintainability, reusability, and performance. Achieving an acceptable architecture requires an iterative derivation and evaluation process that allows refinement based on a series of tradeoffs. Researchers at the University of Texas at Austin are developing a suite of processes and supporting tools to guide architecture derivation from requirements acquisition through system design. The various types of decisions needed for concurrent derivation and evaluation demand a synthesis of evaluation techniques, because no single technique is suitable for all concerns of interest. Two tools in this suite, RARE and ARCADE, cooperate to enable iterative architecture derivation and architecture property evaluation. RARE guides derivation by employing a heuristics knowledge base, and evaluates the resulting architecture by applying static property evaluation based on structural metrics. ARCADE provides dynamic property evaluation leveraging simulation and model-checking. This paper presents a study whereby RARE and ARCADE were employed in the early stages of an industrial project to derive a Domain Reference Architecture (DRA), a high-level architecture capturing domain functionality, data, and timing. The discussion emphasizes early evaluation of performance qualities, and illustrates how ARCADE and RARE cooperate to enable iterative derivation and evaluation. These evaluations influenced DRA refinement as well as subsequent design decisions involving application implementation and computing platform selection.

Architectures embody the requirements expressed by system stakeholders, and the type of architecture used to capture a given set of requirements dictates when evaluation can occur and what will be evaluated. This research aims to leverage requirements and a resulting architecture dictated by the problem domain and captured early in the lifecycle. Thus, the research goal is to provide early performance evaluation in an effort to convey the most accurate blueprint to system implementers specifically and system stakeholders in general. A new software architecture evaluation tool called Arcade, developed to support the Systems Engineering Process Activities (SEPA), automates early performance evaluation of software architectures using simulation. SEPA suggests a comprehensive approach to capture and represent different types of requirements as a multi-level software architecture. One SEPA architecture level, the Domain Reference Architecture (DRA) encompasses performance characteristics inherent to the domain in terms of what processes, data, and timing are required, rather than how a system should be implemented. Performance evaluation of a DRA can provide qualitative data to (1) aid in identification and validation of domain related performance concerns, and (2) provide system implementers with performance guidelines towards satisfying the performance constraints inherent to the domain. The range of performance statistics Arcade is capable of analyzing is demonstrated through a case study of a Motorola e-business project.

When a group comes together to pursue a goal, how should the group interact? Both theory and practice show no single organization always performs best; the best organization depends on context. Therefore, a group should adapt how it interacts to fit the situation. In a multi-agent system (MAS), a decision-making framework (DMF) specifies the allocation of decision-making and action-execution responsibilities for a set of goals among agents within the MAS. Adaptive decision-making frameworks (ADMFs) is the ability to change the DMF, changing which agents are responsible for decision-making and action-execution for a set of goals. Interesting questions remain about the ability of an agent to find the 'best,' or 'near optimal' or 'sufficient' DMF among all the possible DMFs. This paper presents initial exploration of this investigation by asking, 'How does the state space of MAS decision-making organizations scale with growth in the number of agents, number of goals and complexity of the organization structure?' This paper presents tight computational bounds on the size of the search space for DMFs by applying combinatorial mathematics. The DMF representation is also shown to be a factor in the size of this search space.

The benefits of evaluating properties of software architectures stem from two important software architecture roles: (1) providing an opportunity to evaluate requirements and correct defects prior to implementation; and (2) serving as a blueprint for system developers. The paper focuses on a new software architecture evaluation tool called Architecture Analysis Dynamic Environment (Arcade) that uses model checking to provide software architecture safety and liveness evaluation during the requirements gathering and analysis phase. Model checking requires expertise not typically held by systems analysts and software developers. Thus, two barriers to applying model checking must be addressed: (1) translation of the software architecture specification to a form suitable for model checking, and (2) interpretation of the results of model checking. Arcade provides an automated approach to these barriers, allowing model checking of software architectures to be added to the list of techniques available to software analysts and developers focusing on requirements gathering and analysis.

Autonomy is an often cited but rarely agreed upon agent characteristic. Although no definition of agent autonomy is universally accepted, the concept of adaptive autonomy promises increasingly flexible and robust agent-based systems. In general, adaptive autonomy gives agents the ability to seek help for problems or take initiative when otherwise they would be constrained by their design to follow some fixed procedures or rules for interacting with other agents. In order to access these benefits, this article provides a core definition and representation of agent autonomy designed to support the implementation of adaptive agent autonomy. This definition identifies "decision-making control" governing the determination of agent goals and tasks as the key dimension of agent autonomy. In order to gain run-time flexibility and any associated performance improvements, agents must be able to dynamically adapt their autonomy during system operation. This article justifies the implementation of dynamic adaptive autonomy through a series of experiments showing that a multiagent system operating under dynamic adaptive autonomy performs better than a multiagent system operating under fixed autonomy for the same changing run-time conditions.

This paper describes techniques developed for conflict detection during plan integration. Agents' intensions are represented with intended goal structure (IGS) and the E-PERT diagrams. Conflicts are classified as goal, plan, and belief conflicts. Before integrating individual plans and detecting plan conflicts, agents first detect and eliminate their goal conflicts by exchanging their IGS. Plan integration is done through merging individual E-PERT diagrams. Project estimation and review technique (PERT) diagrams have been used extensively in the systems analysis area since the 1980s to provide a global consistent view of parallel activities within a project. We extended the PERT diagrams for use in the plan integration activity within multi-agent systems (MAS). The E-PERT diagram contributes to maintain traceable temporal relations among agents' local scheduled actions. Combined with pattern matching, plan conflicts due to resource sharing, or conflicting conditions (i.e., postconditions of one action disabling preconditions of another action) can be detected. The conflict detection techniques are implemented in a sensible agent testbed to promote deployment and performance analysis.

No abstract available.

This paper discusses a representation for the specification of technology components exploiting the separation of domain requirements (i.e. functional, data and timing requirements) from installation requirements (i.e. implementation specific infrastructure requirements). A process is outlined for building a technology component archive and a set of coefficients is introduced to assist designers in identifying "best-fit" candidates.

Multi-agent systems require adaptability to perform effectively in complex and dynamic environments. This article shows that agents should be able to benefit from dynamically adapting their decision-making frameworks. A decision-making framework describes the set of multi-agent decision-making interactions exercised by members of an agent group in the course of pursuing a goal or set of goals. The decision-making interaction style an agent adopts with respect to other agents influences that agent's degree of autonomy. The article introduces the capability of Dynamic Adaptive Autonomy (DAA), which allows an agent to dynamically modify its autonomy along a defined spectrum (from command-driven to consensus to locally autonomous/master) for each goal it pursues. This article presents one motivation for DAA through experiments showing that the ‘best’ decision-making framework for a group of agents depends not only on the problem domain and pre-defined characteristics of the system, but also on run-time factors that can change during system operation. This result holds regardless of which performance metric is used to define ‘best’. Thus, it is possible for agents to benefit by dynamically adapting their decision-making frameworks to their situation during system operation.

Agent-based systems require flexibility to perform effectively in complex and dynamic environments. Previous research has identified numerous motivations for adaptability in agent-based systems: however, the extent of this adaptability can be expanded. This paper asserts that agents should be able to benefit from controlling the problem-solving frameworks (also called planning-interaction frameworks) under which they plan for their goals. Dynamic Adaptive Autonomy (DAA) allows agents to control their planning-interaction styles, called autonomy levels, along a defined spectrum (from command-driven to consensus to locally autonomous/master). This allows agents to dynamically form, modify, or dissolve goal-oriented problem-solving groups. This paper presents motivation for DAA arguing that the best type of problem-solving framework for a group of agents depends not only on the problem domain and the pre-defined characteristics of the system, but also on run-time factors that can change during system operation. Thus, it is possible for agents to benefit from the capability to dynamically adapt their problem solving framework to their situation.

A multi-agent system (MAS) can be viewed as a group of entities interacting to achieve individual or collective goals. Communication is a central issue in this interaction between agents. This paper uses the COOL language to define an extended version of the contract net protocol to address issues in one particular MAS-sensible agents. The proposed implementation of our protocol and its use in the sensible agent testbed is also discussed.

Finite state machines and their extensions are widely used for system modeling and analysis. Often, the models themselves do not provide much insight into critical system areas which need to be observed (or monitored) for effective control. The issue is mostly left to the discretion of the development team which often has limited domain knowledge. In this paper, an effective method to identify necessary critical system areas that need to be observed and those areas that lead example to the identification of varying levels of control in a design model are discussed. Also, the applicability of this method in differentiating and choosing between two similar system designs is discussed. Extended Statecharts (ESCs) that exploit the XOR configuration of Statecharts for system failure modeling are used for developing the high level system model. The ESC model is subsequently transformed into a high level Petri net model and Petri net analysis techniques are used for the identification of critical system areas for observation and control as well as to differentiate between nearly identical ESC system models.

This paper presents an approach to the assembly sequence planning problem based on a "plan reuse" philosophy. Most of assembly planning research in the past has attempted to completely plan each problem from scratch. This research shows that stored cases of basic assembly configurations can be applied to a given assembly problem. It is observed that the number of such basic assembly configurations is quite small. In the first phase of planning (EVALUATOR phase), the assembly is divided into a number of constituent configurations, which are called "loops". These act as subgoals in its search for solutions (RETRIEVER phase). Plans retrieved for all the subgoals are fused into a set of plans that are consistent with the constraints implied by each plan (MODIFIER and COMPOSITER phases). Application specific constraints on the assembly are explicitly handled in the final phase of planning (POSTPROCESSOR phase).

This paper discusses an architecture designed to provide support for the development of state transition models for an object-oriented distributed environment. The state transition models can be constructed and specified hierarchically as well as derived into new classes of objects through the use of inheritance. A new concept called "exit-safe states" is introduced to assist in the specification of hierarchical state transition models. A graphical monitor to analyze the system at run time has been developed.

This paper presents an approach to the assembly sequence planning problem based on a "plan reuse" philosophy, called the Assembly Planner using Experience (APE). Assembly planning research in the past has attempted to completely plan each problem from scratch. This research shows that stored cases of basic assembly configurations can be applied to a given assembly problem. It is observed that the number of such basic assembly configurations is quite small. Constraints affecting the assembly are also explicitly handled.

This paper studies human-to-human interaction in an attempt to reveal the kinds of human-to-machine interaction that will be necessary for intelligent robot learning of assembly tasks. Experiments were performed in which an "expert" guided an "apprentice" through a complex assembly task using spoken language but no visual communication. An analysis of the dialog reveals that certain protocols and conventions facilitate communication, and that communication breaks down when these protocols are not observed. Five types of protocols were observed: focusing, validators, referencing, descriptors and dialog structure. The implications of these results for human-robot communication are discussed.