The rising cloud-native architecture and adoption of a cloud service provider like Snowflake is significantly increasing the enterprise attack surface in the context of cybersecurity. Snowflake’s cloud data platform provides great scalability and efficiency, but also vulnerabilities to be exploited by malicious actors. While traditional threat detection models have compromised user privacy, in that on-device logs have to be shared with a centralized server, privacy-preserving AI-driven solutions are indeed a necessity. In this paper, we proposed a novel framework that integrates federated learning (FL) and differential privacy (DP) to improve the cyber threat detection in Snowflake environments while keeping the data confidential. This model utilizes secure multiparty computation (SMPC) and homomorphic encryption (HE) for secure data access to minimize the risks of unauthorized access. To this end, we design an AI-based detection framework that ingests cloud telemetry data generated in real time and utilizes privacy-preserving deep learning algorithms to expose advanced cybersecurity attacks. This approach pros is founded on regulatory frameworks (GDPR or CCPA) by balancing accuracy and privacy. We perform exhaustive experiments to assess model effectiveness in terms of detection accuracy, computational efficiency, and privacy preservation trade-offs. Our results show that our approach can better identify zero-day vulnerabilities compared to common ones, all while still preserving strong privacy guarantees. This work has implications for the further development of privacy-aware AI solutions in cybersecurity, leading towards the establishment of secure and resilient cloud computing ecosystems.

Privacy-preserving AI Cyber threat detection Snowflake security Federated learning Differential privacy