Human-machine collaboration requires unambiguous communication to limit misunderstandings. Although semantic interoperability manages to remove ambiguity in machine-to-machine communication, it is insufficient when humans are involved. Humans process and understand information differently based on past experience and the current context, exceeding semantic interoperability’s scope. Cognitive interoperability aims to achieve an aligned understanding, shared intentions, and enable joint decision-making between agents. However, the cognitive state of the human is hard to detect and model, representing a major obstacle to cognitive interoperability. We propose a cognitive Human Digital Twin (cHDT) that emulates a human’s cognitive processes by exploiting cognitive architectures. In particular, we investigate ACT-R as a candidate model. It is a mature cognitive architecture that has been developed based on decades of experimental results from cognitive science and neuroscience. We discuss how the internal state of ACT-R models, and thus the cHDTs, may contribute to cognitive interoperability. With a simplified use case, we illustrate how a cHDT hosting a personalised ACT-R model could track and continuously share the human’s internal cognitive states. This enables external systems, such as robots, to adapt to human perspectives and avoid resource conflicts in human-robot collaboration. Finally, we discuss the applicability of ACT-R as an emulation model, the components of a cHDT, and outline a two-phase implementation plan to validate the proposed solution.