Blockchain technology has emerged as a transformative solution for enhancing transparency, traceability, and immutability in supply chain transactions. However, despite its decentralized security architecture, fraudulent activities such as collusive supplier networks, duplicate invoicing, smart contract exploitation, and phantom shipment generation continue to threaten blockchain-enabled supply ecosystems. Traditional machine learning-based fraud detection models analyze transactions independently and fail to capture the complex relational dependencies inherent in multi-tier supply networks. To address this limitation, this paper proposes a Graph Neural Network (GNN)-based fraud detection framework for blockchain supply networks. The proposed approach models blockchain transactions as graph structures, where nodes represent supply chain entities and edges represent transactional interactions. A Graph Convolutional Network (GCN) is employed to learn structural and feature-bas ed representations of transaction networks, enabling the detection of coordinated and network-level fraudulent behaviors. Experimental evaluation was conducted on a simulated blockchain supply chain dataset comprising 50,000 transaction records and 8,200 interconnected entities. The proposed GNN model achieved an accuracy of 95.2%, precision of 94.6%, recall of 93.8%, and F1-score of 94.2%, outperforming traditional classifiers including Logistic Regression (81.4% accuracy), Random Forest (86.7%), and Artificial Neural Networks (89.3%). Furthermore, the proposed framework reduced false positive rates by 27% compared to baseline methods, demonstrating superior capability in identifying collusive fraud patterns. The results confirm that graph-based deep learning significantly enhances fraud detection performance in decentralized supply chain environments. The proposed system provides a scalable and intelligent security layer for blockchain-enabled supply networks.