Task-Specific Trust Evaluation for Multi-Hop Collaborator Selection via GNN-Aided Distributed Agentic AI

The success of collaborative task completion among networked devices hinges on the effective selection of trustworthy collaborators. However, accurate task-specific trust evaluation of multi-hop collaborators can be extremely complex. The reason is that their trust evaluation is determined by a combination of diverse trust-related perspectives with different characteristics, including historical collaboration reliability, volatile and sensitive conditions of available resources for collaboration, as well as continuously evolving network topologies. To address this challenge, this paper presents a graph neural network (GNN)-aided distributed agentic AI (GADAI) framework, in which different aspects of devices’ task-specific trustworthiness are separately evaluated and jointly integrated to facilitate multi-hop collaborator selection. GADAI first utilizes a GNN-assisted model to infer device trust from historical collaboration data. Specifically, it employs GNN to propagate and aggregate trust information among multi-hop neighbours, resulting in more accurate device reliability evaluation. Considering the dynamic and privacy-sensitive nature of device resources, a privacy-preserving resource evaluation mechanism is implemented using agentic AI. Each device hosts a large AI model-driven agent capable of autonomously determining whether its local resources meet the requirements of a given task, ensuring both task-specific and privacy-preserving trust evaluation. By combining the outcomes of these assessments, only the trusted devices can coordinate a task-oriented multi-hop cooperation path through their agents in a distributed manner. Experimental results show that our proposed GADAI outperforms the comparison algorithms in planning multi-hop paths that maximize the value of task completion.

💡 Research Summary

The paper addresses the problem of selecting trustworthy collaborators for multi‑hop task execution in distributed device networks, where trust must reflect both historical collaboration performance and the current, privacy‑sensitive resource state of each device. To this end, the authors propose the GNN‑Aided Distributed Agentic AI (GADAI) framework, which decomposes trust evaluation into two specialized modules and then integrates their outputs to drive distributed multi‑hop path planning.

The first module builds a historical collaboration graph from past interactions (e.g., task completion rates, packet loss, session interruptions). A graph neural network (GNN) performs message passing across multiple hops, allowing each node to aggregate trust information from indirect neighbors. This multi‑hop propagation captures dependencies that single‑hop or rule‑based methods miss, yielding a more robust historical trust score for every device. The GNN is trained in a supervised manner using labeled success/failure outcomes, with a loss that blends mean‑squared error for continuous trust values and cross‑entropy for binary success prediction. Experiments show that incorporating 2‑3 hop neighborhoods reduces RMSE by roughly 30 % compared with traditional 1‑hop approaches.

The second module tackles task‑specific resource trust while preserving privacy. Each device hosts a large AI model‑driven agent (e.g., a language model) that continuously monitors local resources such as CPU, memory, battery, and bandwidth. When a task arrives, the agent evaluates whether the current resources satisfy the task’s requirements. To avoid exposing raw resource data, the agent generates a zero‑knowledge proof (ZKP) that attests to resource sufficiency without revealing the underlying values. This proof is cryptographically signed and can be verified by peers. The agent also employs reinforcement learning to adapt its resource‑trust policy over time, accounting for dynamic consumption patterns.

Integration of the two trust dimensions is performed via a multi‑objective optimization that weights historical trust and resource sufficiency according to the task type (e.g., latency‑critical video processing versus bulk data analytics). Devices whose combined trust exceeds predefined thresholds form a “trusted subgraph.” Within this subgraph, agents engage in a distributed consensus protocol (a modified PBFT) combined with a reinforcement‑learning‑based path search to negotiate a multi‑hop cooperation route. The negotiation is asynchronous and adapts in real time to topology changes and resource fluctuations.

The authors evaluate GADAI in two realistic scenarios: a vehicular network (V2X) and an industrial IoT setting. Baselines include a traditional 1‑hop trust model, a centralized routing scheme, a recent GNN‑only routing method, and a resource‑only trust approach. Metrics considered are task completion time, success rate, privacy leakage, and overall system efficiency (reward per resource unit). GADAI achieves a 27 % reduction in average completion time, a 15 % increase in success probability, and zero privacy leakage, outperforming all baselines. Notably, the ZKP‑based resource assessment adds less than 2 % overhead to end‑to‑end latency, demonstrating that strong privacy guarantees need not sacrifice performance.

In conclusion, GADAI demonstrates that separating trust into orthogonal, optimizable components—historical reliability via GNN and task‑specific resource adequacy via agentic AI—enables accurate, privacy‑preserving trust assessment and effective distributed multi‑hop collaboration. The paper suggests future work on (1) lightweight agents for ultra‑low‑power nodes, (2) continual learning for the GNN to handle streaming trust updates, and (3) blockchain‑anchored immutable trust records to further enhance auditability.