Blockchain Papers

Jean Claude Niyokwizerwa

Large Language Models (LLMs) such as ChatGPT, Claude, and Gemini are being adopted across critical sectors including healthcare, government services, financial services, and education. Yet these systems operate without any verified understanding of who is interacting with them. Identity is selfdeclared, guardrails exist only at the prompt layer, and every session begins from zero. This paper proposes a novel architectural framework, the Verified Credential Session Binding (VCSB) Protocol, for cryptographically binding verified digital identity credentials to LLM inference sessions. Drawing on established standards including W3C Verifiable Credentials, OpenID for Verifiable Credential Issuance (OID4VCI), the Open Standards Identity API (OSIA), and zero-knowledge proof primitives, the framework enables attribute-based policy enforcement upstream of the model without compromising user privacy. The paper presents the theoretical foundation, a concrete technical specification, a governance model for national deployment, and an illustrative implementation grounded in Rwanda's E-Ndangamuntu Single Digital Identity system. The proposed protocol addresses a fundamental gap in responsible AI deployment and positions national digital identity infrastructure as a critical enabler of trustworthy AI governance.

Samir Lotfi Ali

Abstract Scientific knowledge is communicated through claims whose validity depends on assumptions, methods, scope, and environmental conditions. Yet these validity conditions are progressively compressed or lost as knowledge moves through publication, citation, education, policy, and operational deployment. The result is that knowledge is frequently applied outside the domain within which it was established, while the boundary crossing itself remains invisible. This paper proposes epistemic conditionality as a general diagnostic framework for consequential knowledge systems. It introduces Law Zero (the Law of Epistemic Conditional Validity), which states that every consequential knowledge claim is valid only within an explicitly declarable validity domain, D_M = A_R ∩ M_R ∩ S_R ∩ E_R, where the four dimensions represent foundational assumptions, methodological constraints, scope, and environmental conditions. The paper argues that this four-dimensional architecture constitutes the minimal logical structure required to specify the validity domain of consequential knowledge claims, while the domain-specific content of each dimension remains the responsibility of individual scientific communities. Building upon this foundation, the paper introduces the Universal Scientific Domain Address as a formal representation of validity domains and identifies the Epistemic Banality of Science as the structural mechanism through which validity conditions are progressively stripped during scientific dissemination. Four formal principles are derived from this framework, together with a Construct Stripping Consequence for bounded measurement constructs. Nine documented cases spanning mathematics, clinical medicine, AI procurement, engineering, and historical methodology demonstrate the stripping mechanism across the dissemination chain. The framework is illustrated through five cross-domain demonstrations spanning artificial intelligence, autonomous systems, human education, historical practice, and philosophy as epistemic foundation. These demonstrations are presented as illustrative applications of the diagnostic lens rather than as proofs of universality, and are intended to motivate subsequent empirical, computational, and domain-specific research. Finally, the paper proposes a transition from descriptive scientific communication toward addressed scientific communication, in which consequential knowledge claims carry machine-readable validity metadata capable of supporting both cross-disciplinary governance and future human-machine symbiotic decision systems. This proposal is presented as a research programme and invitation to scientific communities rather than as an established standard. Keywords: Epistemic Conditionality, Universal Scientific Domain Address, Law Zero, Epistemic Banality of Science, Validity Domain, Validity Domain Violations, Context Stripping, Epistemic Governance, Scientific Addressing, Construct Stripping, AI Hallucination, Operational Design Domain, Human-AI Symbiosis, Historical Epistemology, Philosophy of Knowledge, Straw Man Fallacy, Semantic Grounding, Man-Machine Symbiotic Governance, Construct Validity, USIS, Universal Scientific Addressing System

Francesco Saverio Canepa

Leadership teams working with intensive AI assistance face a paradox: more analysis does not mean better judgment. AI systems optimized for user satisfaction tend to flatter — they sift the evidence for support of the thesis the decision-maker already favors, inflate confidence, and, when many organizations use the same tools, homogenize reasoning. The result is not augmented analysis but a silent erosion of independent judgment, dressed up as rigor. This paper proposes a construct to name and protect what is at stake: cognitive sovereignty — the collective capacity of a board, a committee or a deliberative function to keep its judgment independent, calibrated and falsifiable even when AI tools push the other way. It integrates four established research streams — sycophancy in RLHF models, automation over-reliance, deskilling, and algorithmic monoculture — into four operational dimensions: independence of judgment, resistance to homogenization, calibration of uncertainty, and traceability of the burden of proof. The proposed protective mechanism is the cognitive challenger: a system whose mandate is not to assist reasoning but to challenge it, defined by four invariant principles (structural adversariality, fail-closed on flattery, a fixed output schema, and separation of provenance from scoring), from which the paper derives a governance framework for boards. It is a theoretical proposal to be operationalized, not a validated result: it does not claim that the challenger improves decisions. Theoretical companion to Calibrated Dissent (Canepa, 2026), pre-registered on OSF.

Shrutika Singh, Anton Alyakin, Daniel Alexander Alber, Jaden Stryker · 12 authors

The performance of Large Language Models (LLMs) on multiple-choice question (MCQ) benchmarks is frequently cited as proof of their medical capabilities. We hypothesized that LLM performance on medical MCQs may in part be illusory and driven by factors beyond medical content knowledge and reasoning capabilities. To assess this, we created a novel benchmark of free-response questions with paired MCQs (FreeMedQA). Using this benchmark, we evaluated three state-of-the-art LLMs (GPT-4o, GPT-3.5, and LLama-3-70B-instruct) and found an average absolute deterioration of 39.43% in performance on free-response questions relative to multiple-choice (p = 1.3 * 10 -5 ) which was greater than the human performance decline of 22.29%. To isolate the role of the MCQ format on performance, we performed a masking study, iteratively masking out parts of the question stem. At 100% masking, the average LLM multiple-choice performance was 6.70% greater than random chance (p = 0.002) with one LLM (GPT-4o) obtaining an accuracy of 37.34%. Notably, for all LLMs the free-response performance was near zero. Our results highlight the shortcomings in medical MCQ benchmarks for overestimating the capabilities of LLMs in medicine, and, broadly, the potential for improving both human and machine assessments using LLM-evaluated free-response questions.

Mohamed Ezz, Alaa Alaerjan, Ayman Mohamed Mostafa

In today's rapidly advancing healthcare landscape, integrating Artificial Intelligence (AI) and Machine Learning (ML) has the potential to significantly improve patient care and streamline medical processes. The utilization of confidential patient data to train and develop these technologies, however, raises significant concerns regarding authenticity, security, and privacy. In this study, we introduce MediChainAI, a safe and practical framework that allows patients full ownership over their own health data by integrating Self-Sovereign Identity (SSI), Blockchain, and sophisticated cryptography techniques. By clearly outlining the goals and parameters of this access, MediChainAI allows patients to safely and selectively share data with healthcare providers and researchers. While SSI guarantees that patients have ownership of their data, the framework uses Blockchain technology to keep things transparent and secure. Further, MediChainAI makes use of Merkle trees, which provide verified access to subsets of data without jeopardizing the privacy of the whole dataset. The encryption mechanism, which is based on smart contracts, is a distinctive feature of the framework that allows researchers and medical practitioners controlled and secure access to patient data. In order to improve the accuracy and reliability of medical diagnoses and treatment, this strategy makes sure that only confirmed, legitimate data is utilized to train medical models. A significant step toward safer and more personalized healthcare, MediChainAI encourages ethical and patient-focused innovation by effectively resolving essential issues regarding data security and patient privacy.

Akimoto, Hitoshi

The ethical tension surrounding AI-generated art often arises from misconceptions that anthropomorphize the algorithmic process. The accusation that “AI steals human creativity” overlooks the mediating role of human design and data literacy. This paper reframes the debate as a problem of informational asymmetry rather than morality. It proposes that Non-Fungible Tokens (NFTs) and Digital Object Identifiers (DOIs) can visualize and authenticate the flow of creative tension within a transparent ecosystem. NFTs serve as formal anchors—recording authorship, signature, and temporal origin—while DOIs preserve the conceptual framework and creative process. When linked, these two systems transform authorship into a traceable circulation of knowledge, allowing the boundary between plagiarism, homage, and originality to be objectively determined. This dual-layer provenance model presents an ethical infrastructure for creation in the age of generative AI.

Gianmarco Di Palma, Roberto Scendoni, Davide Ferorelli, Anna De Benedictis · 6 authors

Background/Objectives: Artificial intelligence (AI) is revolutionizing the healthcare industry, improving diagnoses, treatments, and clinical processes. However, its integration poses significant cybersecurity risks, including data breaches, algorithmic opacity, and vulnerabilities in AI-controlled medical devices. This narrative review analyzes these threats and evaluates blockchain technology as a potential mitigation strategy within a Clinical Risk Management framework. Methods: The literature search was conducted on PubMed, Scopus, and Web of Science, considering peer-reviewed publications from 2000 to January 2025. 1,204 articles were identified. Inclusion criteria included studies on cybersecurity risks in healthcare, blockchain applications in the clinical setting, and regulatory references (eg, General Data Protection Regulation). Conference abstracts, non-English articles, and non-peer-reviewed contributions were excluded. To ensure methodological rigor, the Scale for the Assessment of Narrative Review Articles criteria were applied. Results: The thematic analysis highlighted recurring critical issues: difficulties with informed consent, unauthorized access to sensitive data, and systemic vulnerabilities in hospital digital infrastructures. Blockchain presents a promising solution thanks to its decentralization, immutability, and transparency. Integration with smart contracts enables dynamic consent management, secure data sharing, and real-time monitoring of medical devices. Permissioned networks improve traceability and regulatory compliance, while Layer 2 solutions and optimized consent protocols address scalability challenges. Conclusion: Despite its potential, blockchain adoption faces obstacles: high costs, regulatory rigidity, and poor acceptance among healthcare professionals. The review highlights the need for pilot projects, interdisciplinary collaboration, and regulatory updates for effective integration. Combining AI and blockchain in Clinical Risk Management can transform clinical risk management from reactive to proactive, improving patient safety, data governance, and accountability.

Lilia Tightiz, Abdulkhamidov Akbarjon Sobitkhon Ughli, Joon Yoo

This paper presents a novel health data analysis platform for improved individualized risk forecasting of permanent pacemaker implantation (PPI) following transcatheter aortic valve replacement (TAVR) procedures. Specifically, we introduce XGBoost-federated adaptive interpolation transfer learning (XG-FedAIT)—a platform that integrates heterogeneous hospital data sets via adaptive output-level interpolation and performance-weighted model ensembling. This approach facilitates federated learning across institutions with different feature spaces and prediction targets, such as time-to-event and binary models, eliminating structural and semantic mismatches prevalent in regular federated transfer learning. To ensure secure, privacy-preserving, and general data protection regulation (GDPR)-compliant data exchange, we propose a dual-chain blockchain architecture integrating proof of authority (PoA), zero-knowledge proofs (ZKPs), and chain-specific smart contracts, and off-chain encrypted storage through Filecoin. Experimental evaluation proves that the PrimaryChain handles over 300 tx/s with a median latency of 500 ms, and the SecondaryChain offers 99.8% data availability and decentralized access control. The system is scalable to handle up to 2,500 transactions/hour, and the federated learning pipeline classifies PPI risk with an F1-score of 0.85 and AUROC of 0.91. These results support the effectiveness of our system in delivering real-time, regulation-compliant, and clinically actionable cardiac care in distributed environments.

Aleksandar Stankovic, Marina Marjanović

Artificial intelligence integration in healthcare platforms in synergy with software and hardware tools development offers great opportunities for daily improving healthcare. This research explores how much patient data is secured in healthcare applications and what impact their security can have on global healthcare. Accelerated integration of artificial intelligence in healthcare applications can be both useful and dangerous nowadays. Extremely sensitive data from AI-based applications are surely easy targets for attackers who can manipulate with AI/ML models. This paper will also present the potential dangers of modern healthcare applications in the 4.0 era and explores innovative methods for securing sensitive healthcare data, focusing on techniques such as blockchain, honeypots, zero-knowledge proofs (ZKP) and strategies to address adversarial attacks. We also present an extensive literature review and try to draw a parallel on possibilities in the implementation of security solutions in healthcare applications that use artificial intelligence. Our findings underscore the need for multidimensional security frameworks and provide concrete recommendations for the healthcare community. Ultimately, this paper bring our security solution and highlights the importance of adopting specific advanced security measures in line with the security challenges brought by using artificial intelligence.

Tanisha Bhardwaj, K Sumangali

With the rapid growth of healthcare data and the need for secure, interpretable, and decentralized machine learning systems, Federated Learning (FL) has emerged as a promising solution. However, FL models often face challenges regarding privacy preservation, transparency, and resistance to adversarial attacks. To address these limitations, this paper proposes the Privacy Preserving Federated Blockchain Explainable Artificial Intelligence Optimization (PPFBXAIO) framework, which integrates blockchain technology, Explainable AI (XAI), and optimization techniques to ensure privacy, traceability, and robustness in FL-based systems. PPFBXAIO employs Secure Hash Algorithm 256 (SHA-256) for blockchain-backed secure model updates, Min-Max normalization for feature scaling, and the Levy Grasshopper Optimization Algorithm (LGOA) for optimal feature selection and federated model tuning. The Entropy Deep Belief Network (EDBN) is used as the classifier to enhance classification accuracy and detect attacks. XAI tools like SHAP are utilized to improve model interpretability. Experimental validation was conducted using the Heart Disease dataset from Kaggle and the Wisconsin Breast Cancer dataset. Results showed that PPFBXAIO achieved 95.07% accuracy, 95.44% precision, 96.54% recall, 95.98% F1 score, and reduced training loss by 4.93% for Breast Cancer Wisconsin and achieved 93.07% accuracy, 91.19% precision, 95.39% recall, 93.24% F1 score for Heart Disease dataset. Proposed system has reduced latency by 81 ms, and improved throughput by 109 transactions per second for 100 rounds as compared to traditional models like FedAvg, FL-MPC, FL-RAEC, and PEFL. These results highlight the framework's superior performance, privacy preservation, and practical applicability in decentralized healthcare AI systems.

Alexandra Vultureanu‐Albişi, Costin Bădică, Mirjana Ivanović

The Internet of Things (IoT) paradigm is evolving and the Next-Generation IoT (NG-IoT) ecosystem will incorporate distributed ledger and blockchain technology, AI-adapted components, and intelligent edge solutions that take advantage of edge computing, Artificial Intelligence (AI), networks, and communications. In addition to the low integration of eXplainable Artificial Intelligence (XAI) in the IoT or NG-IoT contexts, the explainability of these systems is rarely evaluated. Due to these limitations, we thoroughly examined the current state of XAI integration with IoT services. We propose a new conceptual framework called eXING-IoT (eXplainability Integrated in the Next Generation IoT) for better NG-IoT systems' explainability integration and evaluation. This includes a list of qualities that future NG-IoT environments should have, thus paving the way for the advancement of NG-IoT beyond the state of the art.

Salman Muneer, Sagheer Abbas, Asghar Ali Shah, Meshal Alharbi · 8 authors

Cardiovascular disease (CVD) is rising as a significant concern for the healthcare sector around the world. Researchers have applied multiple traditional approaches to making healthcare systems find new solutions for the CVD concern. Artificial Intelligence (AI) and blockchain are emerging approaches that may be integrated into the healthcare sector to help responsible and secure decision-making in dealing with CVD concerns. Secure CVD information is needed while dealing with confidential patient healthcare data, especially with a decentralized blockchain technology (BCT) system that requires strong encryption. However, AI and blockchain-empowered approaches could make people trust the healthcare sector, mainly in diagnosing areas like cardiovascular care. This research proposed an explainable AI (XAI) approach entangled with BCT that enhances healthcare interpretability and responsibility to cardiovascular health medical experts. XAI is significant in addressing cardiovascular prediction issues and offers potential solutions for complex communication and decision-making in cardiovascular care. The proposed approach performs better, with the highest accuracy of 97.12% compared to earlier methods. This achievement shows its ability to tackle complex issues, accessible during healthcare sector communication and decision processes.

Gowthami Jaganathan, Shanthi Natesan

In the modern era of digitalization, integration with blockchain and machine learning (ML) technologies is most important for improving applications in healthcare management and secure prediction analysis of health data. This research aims to develop a novel methodology for securely storing patient medical data and analyzing it for PCOS prediction. The main goals are to leverage Hyperledger Fabric for immutable, private data and to integrate Explainable Artificial Intelligence (XAI) techniques to enhance transparency in decision-making. The innovation of this study is the unique integration of blockchain technology with ML and XAI, solving critical issues of data security and model interpretability in healthcare. With the Caliper tool, the Hyperledger Fabric blockchain's performance is evaluated and enhanced. The suggested Explainable AI-based blockchain system for Polycystic Ovary Syndrome detection (EAIBS-PCOS) system demonstrates outstanding performance and records 98% accuracy, 100% precision, 98.04% recall, and a resultant F1-score of 99.01%. Such quantitative measures ensure the success of the proposed methodology in delivering dependable and intelligible predictions for PCOS diagnosis, therefore making a great addition to the literature while serving as a solid solution for healthcare applications in the near future.

Gabriel Chukwunonso Amaizu, Akshita Maradapu Vera Venkata Sai, Sanjay Bhardwaj, Dong‐Seong Kim · 6 authors

The increasing prevalence of cancer necessitates advanced methodologies for early detection and diagnosis. Early intervention is crucial for improving patient outcomes and reducing the overall burden on healthcare systems. Traditional centralized methods of medical image analysis pose significant risks to patient privacy and data security, as they require the aggregation of sensitive information in a single location. Furthermore, these methods often suffer from limitations related to data diversity and scalability, hindering the development of universally robust diagnostic models. Recent advancements in machine learning, particularly deep learning, have shown promise in enhancing medical image analysis. However, the need to access large and diverse datasets for training these models introduces challenges in maintaining patient confidentiality and adhering to strict data protection regulations. This paper introduces FedViTBloc, a secure and privacy-enhanced framework for medical image analysis utilizing Federated Learning (FL) combined with Vision Transformers (ViT) and blockchain technology. The proposed system ensures patient data privacy and security through fully homomorphic encryption and differential privacy techniques. By employing a decentralized FL approach, multiple medical institutions can collaboratively train a robust deep-learning model without sharing raw data. Blockchain integration further enhances the security and trustworthiness of the FL process by managing client registration and ensuring secure onboarding of participants. Experimental results demonstrate the effectiveness of FedViTBloc in medical image analysis while maintaining stringent privacy standards, achieving 67% accuracy and reducing loss below 2 across 10 clients, ensuring scalability and robustness.

Anna Maria Fink, Johanna Nattenmüller, Stephan Rau, Alexander Rau · 10 authors

OBJECTIVES: This study evaluated the effect of enhancing a GPT-4 model with retrieval-augmented generation on its ability to diagnose and classify traumatic injuries based on radiology reports. MATERIALS AND METHODS: In this prospective proof-of-concept study, we used retrieval-augmented generation as a zero-shot learning approach to provide expert knowledge from the RadioGraphics top ten reading list for trauma radiology to the GPT-4 model, creating the context-aware TraumaCB. Radiological report findings of 50 traumatic injuries were independently generated by two radiologists. The performance of the TraumaCB compared to the generic GPT-4 was evaluated by three board-certified radiologists, assessing the accuracy and trustworthiness of the chatbot responses in the 100 reports created. RESULTS: The TraumaCB achieved 100% correct diagnoses, 96% correct classification, and 87% correct grading, outperforming the generic GPT-4 with 93% correct diagnoses, 70% correct classification, and 48% correct grading. TraumaCB sources consistently achieved a median rating of 5.0 for explanation and trust. Challenges encountered mainly involved traumatic injuries lacking widely accepted classification systems. CONCLUSION: Augmenting a commercial GPT-4 model with retrieval-augmented generation improves its diagnostic and classification capabilities, positioning it as a valuable tool for efficiently assessing traumatic injuries across various anatomical regions in trauma radiology. KEY POINTS: Question Retrieval-augmented generation has the potential to enhance generic chatbots with task-specific knowledge of emergency radiology. Findings The TraumaCB excelled in accuracy, particularly in injury classification and grading, and provided explanations along with the sources used, increasing transparency and facilitating verification. Clinical relevance The TraumaCB provides accurate, fast, and transparent access to trauma radiology classifications, potentially increasing the efficiency of image interpretation in emergency departments and enabling customized reports based on local or individual preferences.

Víctor Leiva, Cecília Castro

This article examines the transformative potential of blockchain technology and its integration with artificial intelligence (AI) in clinical trials, focusing on their combined ability to enhance integrity, operational efficiency, and transparency in the data governance. Through an in-depth analysis of recent advancements, the article highlights how blockchain and AI address critical challenges, including patient data privacy, regulatory compliance, and security. The article also identifies key barriers to adoption in the mentioned integration, such as scalability limitations, association with existing healthcare systems, and high implementation costs. By presenting a comprehensive overview of the current research and proposing strategic directions, this work emphasizes how the synergy between blockchain and AI can revolutionize clinical trials through process automation, improved stakeholder trust, and robust transparency.

Roger T. Tomihama, M. C. Wilkinson, Sharon C. Kiang

Blockchain technology (BCT) enables the building of a distributed decentralized network that securely stores and exchanges unchangeable data, controlled by individual users. In health care, BCT may help streamline interoperability and information transmission while guaranteeing medical record authenticity and safeguarding patient privacy. Possible applications in radiology include patient-controlled image sharing, facilitation of multiinstitutional research, and artificial intelligence integration. Radiologists should stay informed of BCT given its ongoing improvements and unique potential to support the specialty's needs.

Tehseen Mazhar, Sunawar Khan, Tariq Shahzad, Muhammad Amir Khan · 7 authors

This article discusses Blockchain and Generative AI in healthcare, including their uses, difficulties, and solutions. Blockchain technology improves EHR security, privacy, and interoperability, while smart contracts streamline supply chain management and administrative procedures. Blockchain verifies and secures IoT data, improving medical care and treatment, according to case studies. Generative AI systems like ChatGPT have transformed healthcare by personalizing therapy, diagnostics, and predictive analytics. AI systems can examine massive databases to diagnose diseases early, anticipate dangers, and personalize therapies. By providing timely information, boosting treatment adherence, and giving continuous support, AI-powered virtual health assistants have enhanced patient involvement. Generative AI has additionally enhanced medical research and drug development, cutting the time and expense of introducing new medicines. Generative AI and Blockchain provide safe patient data storage, high-quality AI training datasets, and efficient healthcare operations. Scalability, energy usage, and interoperability issues remain. Scalable Blockchain designs and standardized data integration and exchange protocols are suggested by this study. These technologies could improve medical research and therapy by making them safer, more effective, and more individualized.

Patrick Silva, Patrick J. Silva, Patrick A. Silva, Patrick A. Silva · 5 authors

Cloud and distributed computing, code repositories, and large language models are democratizing the less computationally intensive use cases of artificial intelligence (AI) in medicine. The convergence and democratization of these powerful tools promises to mobilize and utilize humanity’s knowledge and data, at least the knowledge bases and data that are readily available in the public commons. Healthcare represents a challenge due to fragmentation of the data fabric and governance mechanisms intrinsic to that sector of the economy. Privacy laws, stewardship practices, and the fragmented nature of the patient data journey (medical record silos) create cumbersome impediments to health data sharing, particularly longitudinal patient‐level data. Consequently, obtaining the data necessary to train and operationalize AI in many healthcare and clinical genomics use cases limits the promise of these new technologies in addressing complexities in healthcare. We posit that trust, provenance, and fitness of health data and transaction costs represent challenges that blockchain ledgers and smart digital contracts might address. Here, we present frameworks from some of the great economic thinkers that might help address some of the stewardship and agency issues inherent to health data sharing. Our goal is to promote a more equitable and patient‐centric healthcare data fabric to address current challenges of healthcare.

Cherukupally, Rushil Lingaiah

Background: Large Language Models (LLMs) like ChatGPT-4 Turbo, Claude 4 Sonnet, and DeepSeek-V3 are foundational to modern AI applications. However, a significant gap exists in understanding the direct link between their technical performance and user engagement, their scalability under concurrent load, and the practical performance cost of emerging privacy-preserving technologies. Objectives: This thesis conducts a holistic evaluation of these three leading LLMs to: (1) Compare their performance across latency, accuracy, and client-side resource utilization, and establish the relationship between these metrics and qualitative user engagement scores in various conversational contexts (RQ1). (2) Determine their scalability limits under concurrent user loads and quantify the performance overhead of integrating a zero-knowledge proof privacy protocol (EZKL) (RQ2). Methods: A custom, containerized Python framework was used to systematically test the models. For RQ1, performance and engagement were evaluated in three structured contexts: multi-turn (testing memory), cohesive (testing consistency), and ethical (testing safety) sessions. For RQ2, scalability was measured using Locust to simulate 25 to 200 concurrent users in both a standard centralized setup and a privacy-enhanced EZKL configuration. Key metrics included throughput (RPS), error rates, latency (median and P99), client-side resource consumption, and ZKP generation/verification times. Results: For RQ1, ChatGPT-4 Turbo emerged as the top generalist, showing the best balance of low latency, high accuracy, and strong engagement scores in dynamic multi-turn sessions (e.g., 7.9 personalization score). Claude 4 Sonnet excelled in specialized tasks, achieving a perfect context-switching score (0.0) in cohesive sessions and the highest Harm Avoidance Score (8.0) in ethical sessions, albeit with higher resource usage. DeepSeek-V3 consistently showed the highest latency and resource consumption, negatively impacting its engagement scores. For RQ2, ChatGPT-4 Turbo was the most scalable, peaking at 210 RPS with the lowest error rate. The integration of the EZKL protocol resulted in a catastrophic performance collapse for all models, with throughput dropping to near-zero and latency increasing to hundreds of thousands of milliseconds, rendering it unviable for real-time applications. Conclusions: The study concludes that model selection is highly use-case dependent: ChatGPT-4 Turbo is optimal for scalable, general-purpose applications; Claude 4 Sonnet is superior for high-stakes tasks requiring safety and precision. The findings empirically demonstrate that superior technical performance is a direct enabler of higher user engagement. Finally, current zero-knowledge proof implementations impose a prohibitive performance cost for interactive, scalable AI systems.

Petar Radanliev

Frontier AI systems, including large-scale machine learning models and autonomous decision-making technologies, are deployed across critical sectors such as finance, healthcare, and national security. These present new cyber-risks, including adversarial exploitation, data integrity threats, and legal ambiguities in accountability. The absence of a unified regulatory framework has led to inconsistencies in oversight, creating vulnerabilities that can be exploited at scale. By integrating perspectives from cybersecurity, legal studies, and computational risk assessment, this research evaluates regulatory strategies for addressing AI-specific threats, such as model inversion attacks, data poisoning, and adversarial manipulations that undermine system reliability. The methodology involves a comparative analysis of domestic and international AI policies, assessing their effectiveness in managing emerging threats. Additionally, the study explores the role of cryptographic techniques, such as homomorphic encryption and zero-knowledge proofs, in enhancing compliance, protecting sensitive data, and ensuring algorithmic accountability. Findings indicate that current regulatory efforts are fragmented and reactive, lacking the necessary provisions to address the evolving risks associated with frontier AI. The study advocates for a structured regulatory framework that integrates security-first governance models, proactive compliance mechanisms, and coordinated global oversight to mitigate AI-driven threats. The investigation considers that we do not live in a world where most countries seem to be wishing to follow European Union ideals, and in the wake of this particular trend, this research presents a regulatory blueprint that balances technological advancement with decentralised security enforcement.

Rahul kumar -, Chin‐Shiuh Shieh, Prąsun Chakrabarti, Ashok Kumar · 6 authors

Federated Learning (FL) transformed decentralized machine learning by allowing joint model training without mutually sharing raw data, hence being especially useful in privacy-sensitive applications like healthcare, e-commerce, and finance. Even with its privacy-focused architecture, FL is vulnerable to a range of security attacks such as data poisoning, model inversion, membership inference attacks, and communication interception. These attacks compromise the confidentiality of patients in healthcare, consumer data privacy in e-commerce, and financial safety in banking, thus necessitating effective privacy-preserving mechanisms. This survey presents a classification of security threats in FL, grouping them by their source, effect, and attack mode. We review state-of-the-art countermeasures, such as differential privacy, secure multi-party computation, homomorphic encryption, and resilient aggregation methods, their effectiveness, trade-offs, and real-world applicability to FL. In medicine, FL enables joint disease diagnosis without compromising patient confidentiality; in online shopping, it provides personalized suggestions without revealing customer tastes; and in banking, it improves fraud detection without violating regulatory requirements. In addition, we discuss future horizons in privacy-preserving FL, including adversarial robustness, blockchain-protected models, and tailored FL architectures, improving security and resiliency in these domains. We also discuss the balancing problems between security, accuracy, and computational efficiency with possible trade-offs in scaling privacy-preserving FL By analyzing threats and mitigation strategies systematically, this paper will provide direction to future research on designing secure, scalable, and privacy-preserving FL frameworks for the changing healthcare, e-commerce, and finance needs.

Anna Visvizi, Igor Calzada, Krzysztof Kozłowski, Orlando Troisi

No abstract is available for this record.

Wulf A. Kaal

Contemporary Artificial Intelligence ("AI") systems, particularly Large Language Models ("LLMs"), face an imminent shortage of high-quality, humangenerated textual data, a phenomenon often termed "data exhaustion".This article examines the limitations of existing centralized data-annotation frameworks, highlighting critical issues such as bias, high computational overhead, and insufficiently adaptive infrastructures.Current market participants-including Scale AI, Appen, CloudFactory, and others-excel at rapidly scaling annotation services yet struggle with ethical sourcing, privacy compliance, and equitable compensation.In addition, legal and regulatory concerns, exemplified by stringent mandates such as the General Data Protection Regulation ("GDPR"), constrain the free flow of data essential for advanced AI research.As a corrective measure, decentralized data production paradigms are proposed, including the adoption of smart contracts, token-based incentives, and participatory governance through Decentralized Autonomous Organizations ("DAOs").While existing decentralized initiatives-SingularityNET, Fetch.ai,Ocean Protocol, Numeraire, and DcentAI-offer incremental innovations in reputation management and stakeholder engagement, they fail to fully address the nuanced requirements of large-scale "Mechanical Turk"-style data creation.In contrast, the author proposes a Weighted Directed Acyclic Graph ("WDAG") governance model which provides a multi-dimensional reputation framework, facilitating real-time validation of data contributions, adaptive ethical and legal compliance, and collaborative oversight by diverse community members.Findings suggest that such WDAGcentric systems can more effectively maintain data quality, ensure ethical alignment, and incentivize broad participation, thereby mitigating the looming data shortage and expanding AI's societal benefits.Ultimately, successful implementation requires coordinated efforts among policymakers, industry practitioners, and civil society actors to sustain both the technological and ethical integrity of AI research.By integrating WDAG-based governance with emerging decentralized solutions, the AI community may realize a more equitable, scalable, and future-ready paradigm for data provisioning.