TREM: A Trust‐Aware Coalition Risk Estimation Model for Multi‐Cloud Environments

ABSTRACT Objective Assessing trust and risk in multi‐cloud environments is increasingly critical as providers collaborate to deliver complex services. Recent advances in large language models (LLMs) enable the automatic generation of diverse coalition scenarios, offering richer and less biased test cases than manually curated datasets. Yet existing approaches, often based on static reputation or individual trust scores, fail to capture the relational and structural factors that drive coalition behavior, leaving hidden vulnerabilities unaddressed. Methods To overcome this limitation, we propose TREM, a trust‐aware coalition risk estimation model that integrates coalition‐level features and applies a logistic regression classifier. This choice provides interpretable probabilistic risk scores while remaining computationally efficient, making the model particularly suitable for security‐critical multi‐cloud environments. Results Experiments on both human‐curated and LLM‐generated coalitions show that TREM achieves accuracy above 90%, maintaining efficiency and transparency. Robustness analysis confirms consistent behavior under class imbalance and coalition size variation, demonstrating that TREM generalizes effectively to diverse and edge‐case coalition structures automatically produced by LLMs.