Automated interoperability plays a pivotal role in managing heterogeneous artefacts across software engineering and data-intensive domains. While conventional approaches often depend on manual or rule-based methods, recent advances in machine learning (ML) and artificial intelligence (AI) present promising avenues for automating interoperability tasks such as matching, mapping, and alignment. However, existing surveys rarely provide a systematic account of how ML/AI techniques address structural and semantic heterogeneity. This paper presents a tertiary study and a critical synthesis of recent developments in AI-driven model interoperability. From a corpus of 82 contributions identified in 19 secondary studies, we selected and classified 19 primary papers that explicitly apply ML/AI methods. Using an extended feature model, we analyze these works along multiple dimensions, including artefact management, execution context, and learning techniques. We also identify key research challenges in explainability, generalization, transformation semantics, and conformance validation, offering a foundation for future research and tool development.