Description of the Invention The present invention relates to the field of cybersecurity and artificial intelligence, and more particularly to a method and system for local autonomous internal penetration testing using artificial intelligence agents augmented by retrieval-based knowledge mechanisms. The invention discloses a technical solution in which one or more autonomous artificial intelligence agents operate within an internal enterprise environment to continuously assess the security posture of information systems, networks, services, and configurations. Unlike conventional penetration testing approaches that rely on manual effort, predefined scripts, or static rule-based scanners, the proposed system dynamically adapts its behavior based on observed system states, retrieved security knowledge, and learned experience. The system employs reinforcement learning to model penetration testing as a sequential decision-making problem, wherein an agent observes an environment state, selects an action from a defined action space, receives a reward based on the outcome, and updates its policy to maximize an expected cumulative reward associated with discovering security weaknesses, misconfigurations, or policy violations. The environment may include hosts, network topology, authentication mechanisms, access control rules, and security monitoring components. To overcome the limitations of fixed context windows and static knowledge in machine learning models, the invention integrates retrieval-augmented mechanisms. At each decision step, the agent retrieves relevant contextual information from a locally stored vector database containing embeddings of security documentation, vulnerability descriptions, configuration policies, compliance requirements, historical findings, and system metadata. The retrieval process is performed using approximate nearest neighbor search, enabling low-latency access to relevant knowledge at scale. The retrieved contextual data is used to augment the agent’s internal state representation and, in some embodiments, to construct augmented prompts for a local large language model responsible for high-level planning, reasoning, or task decomposition. This combination enables the agent to make informed decisions grounded in both learned behavior and up-to-date domain-specific knowledge. In certain embodiments, the system supports multiple cooperating agents, each specializing in a subset of penetration testing activities, such as reconnaissance, vulnerability identification, exploitation simulation, lateral movement analysis, or compliance auditing. Coordination between agents is achieved through a centralized planner or critic during training, while execution may occur in a decentralized manner. A key feature of the invention is that all data processing, retrieval, learning, and inference are performed locally within the organization’s infrastructure, without transmitting sensitive information to external systems. This design preserves confidentiality, complies with regulatory and organizational privacy requirements, and enables deployment in restricted or high-security environments. The disclosed method and system provide continuous, adaptive, and knowledge-informed internal security assessment, improving the detection of security weaknesses while reducing reliance on manual testing and static tools.
The rapid proliferation of distributed and decentralized IT architectures—ranging from cloud-native microservices and edge computing to blockchain-based ecosystems—has fundamentally eroded the efficacy of traditional, centralized information security governance (ISG). As organizational perimeters dissolve into federated, multi-actor, and ephemeral environments, the hierarchical model of a single governing authority enforcing uniform policy becomes not only obsolete but actively detrimental to system resilience. This research examines the critical governance tension between the necessity for central control to manage aggregate risk and the operational reality of local autonomy required for distributed system performance. By adopting a coordination-theoretic lens, this article conceptualizes security governance not as a static command structure but as a dynamic, distributed coordination problem. The study identifies and analyzes specific mechanisms for aligning security responsibilities, decision rights, and assurance processes across autonomous nodes without relying on a single root of trust or a monolithic control plane. Key insights reveal that effective governance in decentralized environments depends on the implementation of polycentric decision-making frameworks, the utilization of automated policy-as-code enforcement, and the adoption of consensus-based assurance mechanisms. The findings suggest that a shift from "governance by mandate" to "governance by protocol" is essential for securing the next generation of digital infrastructure.
Student Information Management Systems (SIMS) are mission critical to higher learning institutions because they govern admissions, registration, fee status, assessment results, progression, graduation, and alumni verification. Yet conventional centralized SIMS architectures may face persistent challenges: record tampering risk, weak end‑to‑end audit trails, fragmented reconciliation across units and campuses, slow and costly credential verification, and limited interoperability with external verifiers.This study develops a conceptual and technical framework for applying Distributed Ledger Technology (DLT) to strengthen SIMS at the Tanzania Institute of Accountancy (TIA). The framework positions DLT as a trust and interoperability layer rather than a replacement for SIMS. It proposes (i) an architecture that anchors cryptographic proofs on‑chain while keeping personal data off‑chain; (ii) standards based digital credentialing using W3C Verifiable Credentials and Decentralized Identifiers; (iii) governance and compliance controls aligned to Tanzanian data protection and cybercrime regimes; and (iv) an implementation roadmap and evaluation metrics grounded in established information systems theories. To make design trade‑offs concrete, the study includes simulated calculations and figures for event volume, storage growth, verification turnaround time, and risk intensity across rollout phases. The framework provides a practical blueprint for a staged pilot at TIA starting with credential verification and assessment audit‑trail anchoring before scaling to additional workflows.
This technical disclosure describes integration patterns for comprehensive smart contract defence systems. The disclosed architecture includes a behavioural suspicion scoring framework, modular security architecture using the Diamond Standard (EIP-2535), graduated response mechanisms, and cross-contract coordination protocols. The system enables layered defence strategies that adapt to threat levels and coordinate responses across multiple contracts. This document is published as a defensive publication to establish prior art and prevent third parties from obtaining patent protection for similar approaches.
The increasing adoption of distributed ledger technology (DLT) in healthcare promises enhanced data security, integrity, and patient control. This study evaluates the characteristics and security implications within a simulated Electronic Health Record (EHR) network, comprising 1,800 records. The analysis focuses on network access control (Access Policy), data integrity mechanisms (Record Hash), and the incidence of security risks (Malicious Node). The records show a near-even split between 'Read/Write' (50.06%) and 'Read-Only' (49.94%) access policies. Crucially, findings reveal that the presence of a Malicious Node (50.72% of records) is highly uniform across both 'Read-Only' (50.95%) and 'Read/Write' (50.5%) policies, indicating that basic access control alone is ineffective at mitigating security risks within this network. This highlights the need for dynamic, context-aware smart contracts that incorporate behavioural or clinical risk factors for enhanced security.
Xinyu Hou, Yang Lu, Rabimba Karanjai, Lei Xu · 5 authors
Ransomware is still one of the most serious cybersecurity threats. Victims often pay but fail to regain access to their data, while also facing the danger of losing data privacy. These uncertainties heavily shape the attacker-victim dynamics in decision-making. In this paper, we introduce and analyze zkRansomware. This new ransomware model integrates zero-knowledge proofs to enable verifiable data recovery and uses smart contracts to enforce multi-round payments while mitigating the risk of data disclosure and privacy loss. We show that zkRansomware is technically feasible using existing cryptographic and blockchain tools and, perhaps counterintuitively, can align incentives between the attacker and the victim. Finally, we develop a theoretical decision-making framework for zkRansomware that distinguishes it from known ransomware decision models and discusses its implications for ransomware risk analysis and response decision support.
This preprint introduces Risk Oracle, an exploit-intelligence–driven SBOM attestation framework designed to support practical risk gating in CI/CD. The system combines signals from Known Exploited Vulnerabilities (KEV) and Exploit Prediction Scoring System (EPSS) with SBOM-to-vulnerability matching to produce a policy-backed gate decision (e.g., pass / warn / block) while keeping the workflow interpretable and reproducible. A core design goal is bounded disclosure: the producer can commit to full findings and then disclose only a limited subset (e.g., top-K highest-risk issues) suitable for downstream verification, reducing disclosure risk while preserving auditability. The paper details the end-to-end pipeline (producer/verifier roles), a typed attestation schema, a scoring and decision procedure, and an evaluation that studies (i) signal behavior under pinned KEV/EPSS snapshots and (ii) operational overhead under synthetic SBOM scaling intended to approximate CI workloads. Key contributions A practical SBOM attestation pipeline that integrates exploit-intelligence signals for operational decision-making in CI/CD. A typed attestation schema and verifier procedure supporting bounded disclosure. A policy-driven scoring and gating framework (pass/warn/block) grounded in vulnerability-management practice. Evaluation focused on interpretability and operational cost (runtime/payload scaling) under reproducible, pinned snapshots. Artifacts / Reproducibility Code, scripts, and pinned snapshot references: [GitHub link] Suggested citationSudhir, S. (2026). Risk Oracle: Exploit-Intelligence–Driven SBOM Attestations with Bounded Disclosure (preprint). Zenodo. DOI: [10.5281/zenodo.18153487] Keywords: SBOM, software supply chain security, vulnerability management, KEV, EPSS, OSV, attestation, CI/CD, bounded disclosure, reproducibility
Md Al Amin, Indrajit Ray, Indrakshi Ray, Yashwant K. Malaiya · 5 authors
Access to electronic health records (EHRs) is heavily regulated by various policies, including federal-level policies, state-level statutes, international data protection laws, and local and organizational-level policies. These policies may include procedures to ensure compliance with other organizational-level regulations. In addition, individual patients can establish agreements, formally known as patient-provider agreements (PPA), with their healthcare providers to express their consent to access or share their protected health information (PHI). When such policies are adequately specified and implemented, they go a long way toward protecting EHR data. However, research has shown that significant policy compliance problems or gaps often go undetected until after a breach or security incident. Further, a recent study shows that subcultures within a healthcare organization influence whether employees violate policies, perhaps unintentionally. These observations motivate us to revisit the compliance and provenance aspects of policies. This dissertation proposes a blockchain-powered, smart contract-based policy-compliance assurance framework to enforce patient-provider agreements and other applicable policies and attributes, ensuring policy compliance and provenance in the healthcare sector. This work proposes a novel compliance review mechanism, Proof of Compliance (PoC), that conducts reviews through a set of independent, distributed, decentralized auditor nodes from various stakeholders, such as healthcare organizations, insurance companies, federal and other government agencies, regulatory agencies, and others mandated by the business requirements. Blockchain smart contracts appear to be a promising new technology for enforcing policies. In addition, blockchains' immutable storage properties and strong integrity guarantees provide hope that an adequate trail of policy compliance (or non-compliance) can be maintained, thereby facilitating provenance.
Smart contract security research has historically emphasized exploit-driven threat models developed for open, permissionless blockchain environments. While effective for identifying adversarial attacks and loss-of-funds scenarios, these models are insufficient for institutional blockchain systems supporting Real-World Assets (RWAs). In regulated and asset-backed deployments, threats frequently arise from non-adversarial actors, design assumptions, operational dependencies, and compliance enforcement failures rather than from direct malicious exploitation.This paper presents a structured threat taxonomy tailored to institutional RWA smart contract systems. The taxonomy classifies threats according to origin, intent, capability, and impact, explicitly incorporating adversarial, non-adversarial, and systemic threat sources. By decoupling threat analysis from exploit-centric assumptions, the taxonomy enables correctness-oriented risk assessment, formal specification of threat boundaries, and alignment with institutional audit and compliance requirements. The proposed framework provides a foundation for secure system design, verification, and operational governance in regulated blockchain environments.
We report on a systematic measurement study of unauthenticated Ollama AI inference servers exposed on the public internet. Using passive enumeration via the Shodan Internet Intelligence Platform, we identify 20,971 publicly accessible Ollama endpoints and more than 54,000 total AI inference services across more than ten backend types, all reachable with no authentication. We conduct targeted proof-of-concept probing against three representative endpoints spanning enterprise cloud, residential broadband, and commercial VPS infrastructure across three countries (USA, Sweden, Austria). All three endpoints score maximum exploitability across five dangerous write operations (model deletion, arbitrary pull, push exfiltration, persistent model poisoning, and clone operations), zero security headers, and zero rate limiting. We document a previously unreported financial fraud vector arising from the proliferation of cloud API proxy models (:cloud suffix), through which unauthenticated access to an Ollama endpoint constitutes direct financial fraud against the endpoint owner's commercial AI API accounts. Extending beyond prior work, we characterize four Retrieval-Augmented Generation (RAG) layer PII exfiltration vectors that require no write access or special knowledge: (1) context-token session hijacking via /api/generate's persistent KV-cache state; (2) cleartext HTTP conversation history re-transmission; (3) co-deployed RAG database exposure; and (4) prompt-based RAG context extraction. In live testing, the fourth vector extracted 11 of 11 PII categories including financial credentials, medical diagnoses, and authentication passwords on the first plain-language prompt against a live production endpoint. We additionally identify a supply chain risk arising from shared model artifact digests across geographically separated endpoints, and a novel operator risk taxonomy distinguishing technically sophisticated but selectively insecure operators from the commonly assumed unaware operator class. All probing used entirely synthetic personal data. No real user conversations were accessed or retained. Responsible disclosure notices were sent to relevant ISPs prior to submission.
This report studies EvoVuln, a smart contract vulnerability analysis pipeline that represents vulnerability knowledge as a natural-language detection method, refines that knowledge through iterative training, and converts the result into an executable detection plan for downstream contract analysis. The report focuses on the Stage 2 update-acceptance mechanism, which determines whether a candidate knowledge revision should replace the current knowledge state during labelled-data refinement. In the baseline implementation, candidate updates are accepted based only on improvement over the previously misclassified subset. While simple and efficient, this rule may accept revisions that improve local error cases while worsening behaviour on other labelled contracts. To address this limitation, this report investigates a modified acceptance rule that adds full-set validation before acceptance, requiring candidate knowledge to reduce total errors on the full labelled training set. The modified rule was implemented as a localised change at the Stage 2 acceptance point, while the broader pipeline structure was retained. Evaluation was conducted on two selected vulnerability types using matched starting snapshots and repeated full-pipeline runs under both policies. The modified rule consistently accepted fewer candidate updates than the baseline policy, indicating stricter filtering during refinement. It also improved average downstream F1 in both vulnerability types, although the form of improvement differed: for price manipulation, recall increased while precision decreased, whereas for access control, average precision, recall, and F1 all improved. These findings suggest that Stage 2 acceptance is a meaningful design choice in iterative LLM-based vulnerability analysis pipelines. Within the scope of the evaluated cases, adding full-set validation was associated with improved downstream detector performance, although the results should be interpreted cautiously given the limited experimental scope and the pipeline’s sensitivity to LLM-related non-determinism.
Graph neural networks (GNNs) have shown notable success in identifying security vulnerabilities within Ethereum smart contracts by capturing structural relationships encoded in control- and data-flow graphs. Despite their eff... | Find, read and cite all the research you need on Tech Science Press
Decentralized and autonomous systems associated with Web 3.0 challenge long-standing assumptions about security governance, responsibility, and control. Although advances in cryptography, consensus mechanisms, and automation have strengthened technical protections, they have also fragmented accountability across software agents, organizations, and ecosystems, producing security failures in environments where responsibility for prevention, response, and remediation is unclear or contested. This paper demonstrates that many such failures stem not from inadequate technical safeguards but from governance gaps created by sociotechnical complexity. Drawing on sociotechnical systems theory, it introduces the LAG (Layers-Accountability-Governance) framework, which distinguishes among information technology security, information security, and enterprise-level governance, and clarifies the misalignment between ex ante preventive mechanisms and ex post response and recovery processes. Through analysis of decentralized architectures, autonomous agents, and machine identities, and case studies of the DAO, Poly Network, and oracle-related failures, the paper shows how technically correct system behavior can nonetheless produce governance failure and discusses implications for the design and governance of secure systems in complex digital ecosystems where traditional organizational boundaries no longer apply.
The emergence of coordinated, multi-agent offensive systems in cyberspace—variously manifesting as distributed reconnaissance campaigns, AI-assisted vulnerability discovery pipelines, adaptive lateral movement swarms, and cross-domain settlement attacks—has outpaced the formal theoretical treatment necessary for principled defence. This monograph addresses that gap with five primary contributions. First, we model swarm versus counter- swarm interaction as a two-player partially observable stochastic game (POSG) and derive equilibrium conditions under asymmetric information. A sufficient condition is established, via Fano’s inequality, under which an ambiguity-preserving offensive policy becomes dominant on the induced information set; the general exact-solution problem is shown to be intractable unless P = NEXP, because the POSG family strictly contains finite-horizon decentralised partially observable Markov decision processes as a special case. Second, we introduce a formal taxonomy of five autonomous cyber swarm attack classes—distributed reconnaissance, adaptive lateral movement, threshold-splitting exfiltration, semantic service exhaustion, and cross-domain settlement—each coupled to an exact detection decision problem with a worst- case complexity lower bound, an information-theoretic defender error floor, and a formal evasion condition. Third, we propose AEGIS-MESH (Attested Evidence-Gated Interdiction System for Multi-domain Event-Synchronized HotStuff), a new counter-swarm architecture specified as a process-algebraic state machine family, with Byzantine fault tolerance for committees of n ≥ 3f + 1 replicas and five temporal-logic safety invariants enforcing evidence gating, mandatory human approval for destructive actions, and causal auditability. Fourth, we prove that the swarm attribution problem is NP-complete by reduction from Subgraph Isomorphism and derive a closed-form sensor-density lower bound required to guarantee attribution error at most δ. Fifth, we conduct a parametric numerical analysis of the dominance condition, a systematic capability evaluation of contemporary autonomous defence systems against a proposed reproducibility and external-validity framework, and a formal treatment of three Web3 security problems—bridge validator-threshold compromise, blind signing, and TWAP oracle manipulation—before closing with treaty-grade draft governance language and an auditable definition of meaningful human control.
Open access
Infrastructure Resilience and Vulnerability Analysis
This exploratory study investigates the security, fairness and economic implications of Decentralised Finance (DeFi) within the Ethereum ecosystem. While DeFi has experienced rapid growth, there remains several systemic risks and vulnerabilities that can affect hugely. The study focuses on three key aspects: (1) Transaction Semantics (Analysing how core actions in DeFi are executed on EVM and recorded in transaction receipts and logs), (2) Vulnerability Analysis (Identifying recurring patterns of anomalies), and (3) Detection Framework (Evaluating the effectiveness and potential scale of DeepTx, a real-time analysis tool to detect threats in transactions). By conducting the report based on leading protocols in Ethereum such as Uniswap, Aave and Compound, the study bridges the gap between transaction data and security insights. It will therefore provide a foundation for policymakers to create frameworks for how bugs and risks can affect the DeFi ecosystem, serving as a guide for a secure and scalable Web3 industry.
The adoption of decentralized technologies in healthcare introduces new opportunities for secure, patient-centered data management but also brings significant privacy and security challenges. This paper presents a threat modeling approach applied to a Web3-based healthcare platform that integrates blockchain for access logging, a FHIR-compliant server for clinical data, and a backend for identity and access management. Using the LINDDUN privacy threat modeling framework and OWASP Threat Dragon, we identified and prioritized privacy risks based on system architecture and data flows. The results show that threat modeling can provide early insights into regulatory compliance, data exposure, and user privacy concerns. This process can be viewed as a foundational step in the development of digital health systems. While the analysis was focused on a specific use case, the methodology is adaptable to a wide range of applications handling sensitive personal data.
We frame Ethereum transactions reverted by invariants-require(<invariant>)/ assert(<invariant>)/if (<invariant>) revert statements in the contract implementation-as a positive signal of active on-chain defenses. Despite their value, the defensive patterns in these transactions remain undiscovered and underutilized in security research. We present Raven, a framework that aligns reverted transactions to the invariant causing the reversion in the smart contract source code, embeds these invariants using our BERT-based fine-tuned model, and clusters them by semantic intent to mine defensive invariant categories on Ethereum. Evaluated on a sample of 20,000 reverted transactions, Raven achieves cohesive and meaningful clusters of transaction-reverting invariants. Manual expert review of the mined 19 semantic clusters uncovers six new invariant categories absent from existing invariant catalogs, including feature toggles, replay prevention, proof/signature verification, counters, caller-provided slippage thresholds, and allow/ban/bot lists. To demonstrate the practical utility of this invariant catalog mining pipeline, we conduct a case study using one of the newly discovered invariant categories as a fuzzing oracle to detect vulnerabilities in a real-world attack. Raven thus can map Ethereum's successful defenses. These invariant categories enable security researchers to develop analysis tools based on data-driven security oracles extracted from the smart contracts' working defenses.
Proof-of-Stake (PoS) has become a widely adopted low-energy consensus paradigm, yet its incentive mechanism remains vulnerable to strategic deviations under complex temporal and network conditions. Existing analyzes rely heavily on theoretical reasoning or manually crafted scenarios, leading to limited coverage and substantial expert overhead. This paper presents PoSitive, an automated framework for systematically uncovering incentive weaknesses in PoS consensus. PoSitive establishes a closed-loop workflow composed of four cooperative modules: a scenario construction module that generates diverse and controllable adversarial configurations, a scenario execution module that faithfully reproduces validator interactions, an outcome evaluation module that quantifies incentive deviations and consensus instability, and a policy optimization module that employs reinforcement learning to iteratively refine attack strategies and explore a broader strategic space. Experimental results demonstrate that PoSitive can effectively identify incentive-layer security risks. Using this framework, we uncover three previously unknown attack strategies, and comparative experiments further confirm the significant role of the policy optimization module in enhancing both attack quality and success rate.
Information and Cyber Security
Infrastructure Resilience and Vulnerability Analysis
Smart contract vulnerabilities have led to massive losses in digital assets. While researchers have proposed numerous detection methods utilizing static analysis, fuzzing, and deep learning, most are limited to identifying vulnerabilities within individual contracts. Consequently, these approaches fail to effectively analyze cross-contract interactions via external function calls, resulting in false negatives and positives. To address these limitations, we present VulCrosser, a deep learning method tailored for vulnerability detection in contract interaction scenarios. VulCrosser enables comprehensive risk assessment by analyzing function call chain traces. Specifically, it constructs a Cross-Contract Dependency Graph (CCDG) to effectively model inter-contract dependencies, network dynamics, and interaction semantics. It then employs a heterogeneous graph neural network with a two-level attention mechanism to extract and integrate complex features from the graph, ultimately achieving accurate risk assessment. We evaluated VulCrosser on three common vulnerabilities: reentrancy, timestamp dependency, and transaction state dependency. Experimental results show that VulCrosser outperforms all baseline methods, improving detection accuracy by 5.04%, 4.39%, and 5.09%, and $F 1$ scores by $4.93 \%, 4.60 \%$, and 4.99%, respectively.