A Zero Trust Framework with AI-Driven Identity and Intrusion Detection for Multi-Cloud MLOps

Multi-cloud strategies are transforming enterprise MLOps deployments but introduce security risks such as model poisoning, adversarial inference, and insider threats. Traditional IAM solutions with static policies struggle to protect AI -driven workflows. This paper presents an AI-enhanced Zero Trust Secu-rity (ZTS) framework integrating Federated Identity Management (FIM) and an AI-powered Intrusion Detection System (IDS) to secure multi-cloud MLOps environments. The FIM component is validated using a real-world authentication dataset (LANL), while IDS is tested on intrusion datasets (UNSW-NB15, CIC-IDS2018). The FIM module employs Graph Neural Networks (GNNs) and Long Short-Term Memory (LSTM) models for adaptive authentication. High-risk interactions enforce Multi-Factor Authentication (MFA), while low-risk users experience expedited ac-cess. Simultaneously, the AI-driven IDS continuously monitors network traffic and user behavior to detect MLOps-specific threats, including model poisoning and inference API exploitation. Experimental results validate the framework's effectiveness, demonstrating a 38 % reduction in authentication latency, 42 % improvement in threat detection accuracy, and 67 % fewer false positives over baseline methods. The system also scales effectively under increasing user loads. This research highlights the critical role of AI -driven Zero Trust strategies in securing distributed MLOps work-flows, offering a proactive, scalable defense against evolving cyber threats.