In the present study, we propose a rare-event sampling method called anomaly detection parallel cascade selection molecular dynamics (ad-PaCS-MD). The original PaCS-MD was designed to generate conformational transition pathways from a given reactant to a product when the latter is known a priori. As an extension of the original method, ad-PaCS-MD has been designed to efficiently search transition pathways from a given reactant without referring to a given product. In ad-PaCS-MD, rarely occurring but essential states (configurations) of proteins for the transitions are identified based on the degrees of an anomaly. In more detail, ad-PaCS-MD adopts an algorithm called an anomaly detection generative adversarial network (anoGAN) as a measure for detecting rarely occurring states to be resampled. Here, the essential configurations with higher degrees of the anomaly are selected with anoGAN and intensively resampled by restarting short-time MD simulations from the selected configurations. By repeating the detections and resampling of configurations with the higher degrees of the anomaly, ad-PaCS-MD automatically and efficiently promotes the rare events and gives a wide range of the free energy landscape by combining with the Markov state model construction. As demonstrations, open–closed transitions of two globular proteins (T4 lysozyme and maltose-binding protein) were promoted with ad-PaCS-MD by referring only to the given starting configurations. In each demonstration, ad-PaCS-MD promoted the large-amplitude open–closed transitions with nanosecond-order simulation times. In conclusion, our demonstrations showed a higher conformational sampling efficiency for ad-PaCS-MD than conventional MD (CMD) because CMD required computational costs of more than microsecond-order simulation times to promote the rare events.