<![CDATA[Automatic Identification System data generate continuous spatiotemporal information that supports maritime surveillance, yet converting detected vessel anomalies into reliable analytical reports remains challenging because statistical abnormality does not always indicate operational significance. This study aimed to conceptualize and evaluate an Artificial Intelligence-supported mechanism for generating evidence-grounded vessel AIS anomaly analysis reports. A Design Science Research approach was employed through AIS preprocessing, trajectory reconstruction, anomaly detection, contextual interpretation, explainability, evidence packaging, controlled natural-language generation, and expert validation. Findings showed that the anomaly-detection component achieved satisfactory predictive performance, while report quality improved substantially after provenance, contextual validation, and uncertainty controls were incorporated. Expert assessments indicated stronger factual consistency, traceability, contextual relevance, and uncertainty communication in the refined reports, while case analyses demonstrated that AI could distinguish observed AIS facts from model inference more effectively when constrained by structured evidence. The study concludes that generative AI should function as an evidence-constrained analytical communicator rather than an autonomous interpreter of vessel intent or maritime risk. The proposed architecture contributes an integrated framework for transparent, auditable, and human-verified AIS anomaly reporting and provides a foundation for future validation across diverse maritime environments, vessel classes, and operational contexts under realistic human-in-the-loop maritime decision-support conditions across waters.]]>
