The Social Internet of Things (SIoT) is a fast growing technology that autonomously connects things, allowing for exchange of information depending on relationships. However, the open nature of SIoT exposes it to various security challenges, such as default passwords, brute-forcing attacks, data privacy breaches, and an increasing number of IoT botnets. To address these vulnerabilities, a novel operational security framework for SIoT, leveraging the robust properties of blockchain technology. The primary objectives of the operational security model are to establish a secure and trusted environment in SIoT, protect sensitive data, ensure reliable device authentication, and enable secure and encrypted communication in SIoT operations. The proposed model combines key features of blockchain, including decentralization, immutability, transparency, and cryptographic mechanisms, with SIoT security operations. Techniques like encryption, anonymization and differential privacy are employed to preserve privacy in SIoT devices and safeguard sensitive data. An empirical assessment of this operational security framework yields promising outcomes in the pursuit of its defined objectives. It furnishes a decentralized and tamper-proof foundation for device authentication, thereby cementing the trustworthiness of SIoT devices. Moreover, the proposed Graph Neural Network (GNN) model eclipses the performance of traditional models, boasting a 95% accuracy, a 96% precision, a 95% recall, and a 95% F1-score. This achievement can be attributed to the incorporation of computations that encapsulate both transitivity and composability among devices. In tandem, the suggested framework emerges triumphant when compared to other methodologies, spanning diverse metrics such as throughput, latency, security score, privacy score, and trust score. Notably, this framework not only secures communication but also upholds data integrity and confidentiality, ensuring exclusive access to SIoT devices and data by authorized entities. The proposed SIoT framework, with 280 Tx/s throughput, 2.2 seconds latency, 9.5 security score, 9.0 privacy score, and 9.5 trust score, outperforms other authors’ models across all the parameters.