Abstract: Objectives This paper introduces a novel data-driven approach for generating realistic hazardous overtaking scenarios, crucial for rigorously evaluating the autonomous collision avoidance capabilities of intelligent ships. Current methods for generating such scenarios often struggle to balance diversity, realism, and the representation of high-risk situations. Our method addresses this limitation by leveraging the wealth of information contained within Automatic Identification System (AIS) data to synthesize diverse and realistic overtaking encounters. Methods Specifically, we propose a hybrid model that integrates a sequence generative adversarial network (SeqGAN) with a self-attention mechanism. The SeqGAN learns the complex patterns and dynamics inherent in AIS-derived ship trajectories, enabling the generation of novel, yet plausible, overtaking maneuvers. The incorporation of a self-attention mechanism further enhances the model’s ability to capture long-range dependencies within the trajectories, resulting in more realistic and nuanced simulations. To ensure the generated scenarios accurately reflect high-risk situations, we have developed a constraint model that utilizes longitudinal and lateral safety distances between vessels to define realistic initial conditions. This model dynamically adjusts the initial positions and velocities of both the target vessel and the intelligent ship undergoing testing, guaranteeing that each generated scenario presents a genuine collision risk. Results The efficacy of our approach is validated through extensive simulations. We generated 500 high-risk overtaking scenarios, demonstrating a significant improvement in test scenario coverage. Impressively, 97.3% of these generated trajectories fall within a predefined buffer zone encompassing real-world trajectories, confirming the high fidelity of our model. Furthermore, the speed distributions of the generated target vessels closely match those observed in real-world AIS data, further corroborating the realism of our approach. Conclusions The enhanced realism and diversity of scenarios generated by this method significantly improve the efficiency of autonomous collision avoidance testing. This allows for a more precise definition of safety performance boundaries, accelerating the development and refinement of autonomous collision avoidance algorithms. Ultimately, this contribution facilitates the creation of safer and more reliable intelligent shipping capable of navigating the complexities of modern maritime environments.