A Novel Method to Mitigate Adversarial Attacks Against AI-as-a-Service Functionality

Artificial Intelligence (AI) will play a crucial role in enabling 6G technology. In comparison to current networks that mainly offer ubiquitous communication capabilities, 6G aims to transform the network into a robust, distributed AI platform, making its AI capabilities accessible to consumers. This will be facilitated by the AI-as-a-Service (AIaaS) framework, which allows network service providers to effectively utilize AI as a service, opening up new possibilities. Therefore, it is vital to ensure that the AIaaS framework is well-protected against malicious attacks. However, research on Black-Box attacks indicates that merely having access to the model’s input and output is sufficient to perform adversarial attacks by creating perturbations that can deceive AI models. In this study, we propose an effective defense mechanism to mitigate inference query-based Black-Box attacks which might target AI models deployed in an AIaaS framework. We have experimentally evaluated and verified the efficacy of our approach on standard datasets.