An AI model denial-of-service attack occurs when an attacker intentionally submits very large or computationally expensive inputs that force the model to consume excessive processing resources. This overwhelms the model's inference capacity, degrades performance, and can prevent legitimate users from receiving responses, effectively disrupting the availability of the AI service.