fine-grained trust management

本文通过整合可信AI治理、代理安全等方法，提出信任即服务（TaaS）来解决跨组织边界行动的信任评估问题。

This study addresses the lack of systematic research and a unified framework concerning trust in digital twin systems. Through a systematic literature review and content mapping, complemented by a qualitative analytical framework, the work categorizes and synthesizes trust challenges and enhancement strategies reported in existing review literature. It identifies seven core trust challenges along with their corresponding mitigation strategies, proposes four types of trust integration models, and reveals four distinct trust-prioritization paradigms: human-centric, safety-critical, context-specific, and technology-driven. Innovatively, the study advocates for emerging directions such as trust-by-design, embedding trust metadata, and examining architectural impacts on trust, thereby establishing a theoretical foundation and research roadmap for developing trustworthy digital twin systems.

This work addresses the lack of a unified trust mechanism across data, service, and knowledge layers in industrial intelligence, which undermines reliability, accountability, and regulatory compliance. To bridge this gap, the authors propose Trisk—a novel framework that establishes the first holistic view of cross-layer trustworthy governance in industrial settings. Trisk introduces a five-dimensional trust model integrating quality, safety, privacy, fairness, and explainability, supported by a complementary three-dimensional classification scheme. By synergistically combining enabling technologies—including knowledge graphs, zero-trust architectures, causal reasoning, and federated/edge computing—the framework systematically addresses critical gaps in semantic interoperability and policy enforcement. Grounded in a comprehensive review and maturity assessment of over 120 studies, this research provides a systematic roadmap for advancing both theoretical foundations and practical deployment of trustworthy AI in Industry 5.0.

Existing Byzantine fault-tolerant CRDT systems lack evolvable and replicable state management for trust relationships and governance rules. This work proposes a dual-CRDT architecture: a Trust CRDT models governance policies and trust relations as dynamically evolvable CRDT states, while a Data CRDT deterministically reconstructs its state based on this trust configuration, enabling co-evolution of governance and data. By incorporating governance itself into the CRDT state, the design introduces a recursive trust governance model augmented with a Byzantine trust filtering mechanism. A prototype implementation on the Melda/melda-sec platform demonstrates the feasibility of self-consistent, evolvable trust governance in decentralized environments.

This work addresses the challenge of securely excluding historical updates from compromised nodes in traditional Byzantine fault-tolerant CRDTs without violating causal consistency. The authors propose a fine-grained trust model that, for the first time in Byzantine CRDTs, decouples identity trust from content trust. By integrating deterministic reconstruction, public-key-based identity verification, and a semantics-aware update filtering mechanism, the approach enables selective inclusion or exclusion of updates. This design supports application-level policies and effectively mitigates Byzantine behavior and faulty nodes while strictly preserving causal consistency, thereby significantly enhancing the robustness and flexibility of decentralized systems in post-compromise scenarios.

本文提出了一种针对代理过程的证据声明模型，以解决日志和锚点在语义上的误导问题，通过定义多种证据声明来提高AI系统的可信度。

RepuLink通过两层信任与声誉模型及后向背书奖惩传播机制，解决了分布式网络中责任归属和新节点初始信誉问题。

This work addresses the challenge of ensuring safety, reliability, and trustworthiness in collective adaptive systems operating in dynamic environments by proposing a modular design paradigm centered on intrinsic trustworthiness. The approach integrates a runtime model based on local causal event sequences, a temporal logic verification technique supporting modular architectures, and a compositional reasoning mechanism for global system properties grounded in component attributes. Through this tripartite framework, the study overcomes key limitations of conventional formal methods and demonstrates substantial improvements in verifiability and scalability in case studies, thereby establishing both a theoretical foundation and a practical pathway for engineering highly trustworthy collective adaptive systems.

This study addresses the security risks arising from semantic mismatches in data that crosses trust boundaries, even when such data passes syntactic validation. It introduces the first systematic definition of the “Trust Boundary Semantic Gap” (TBSG) and proposes a Multidimensional Trust Boundary Semantic Gap (MDTBSG) model that characterizes TBSG along four dimensions: identity, space, time, and interpretation. Furthermore, the work develops the TBSAM framework for the design phase, integrating static specification analysis, semantic alignment modeling, gap provenance tracing, and architectural control mapping to identify, prioritize, and mitigate semantic gaps. Applied retrospectively to the SolarWinds/SUNBURST attack, the approach successfully pinpointed the root cause of critical semantic gaps, clarified assumptions in the receiving domain, and recommended effective architectural controls to disrupt the attack path.

This work addresses the lack of a unified trust visibility mechanism in cross-vendor AI agent tooling, which can lead to cascading failures when compromised tools are repeatedly invoked. To mitigate this, the paper proposes AgentToolMO, a framework grounded in the 3GPP NRM information model that integrates a formal trust state machine and a bounded-convergence damped cascade propagation mechanism. Trust states across vendors are disseminated in near real time and their impacts traced via standardized MnS interfaces. Experimental results demonstrate that the approach reduces failure response latency from hours to near real time, guarantees cascade convergence within a finite number of iterations, and incurs sublinear notification overhead, thereby significantly enhancing system resilience and interoperability.