Processes for detecting, responding to, and recovering from AI system failures or security events. AI incident response plans often address model-specific scenarios such as hallucination-caused harm, prompt injection breaches, and unexpected behavior changes; contracts commonly specify notification and cooperation obligations.