Recent, rapid information technology advancements have led to numerous AI models being applied in society, with innumerable AI applications across various fields. However, as AI models become central to societal functions, they inevitably attract the attention of malicious actors, resulting in the proliferation of counterfeit AI models. Therefore, identity verification, like user recognition, is essential for AI models as well. However, due to probabilistic variability in outputs, especially in large language models (LLMs), and continuous capability enhancement through autonomous learning, the typical user recognition method where the same information is enrolled and presented cannot be directly applied. To address this, in this study, we proposed and investigated an AI model identification method capable of handling the inherent variability in AI outputs while accurately verifying the authenticity of AI models whose capabilities continuously evolve. Specifically, we defined the requirements for AI model identification by “anthropomorphizing AI models” and developed a method based on “assessing the abilities of AI models.” The AI model identification method, designed based on these two concepts, is termed the Completely Automated Public Test to Tell Ability of Artificial Intelligence (CAPT-AI).