An AI incident is a harmful or significant event involving an AI system that needs logging, review, and sometimes escalation. It can include unsafe outputs, model drift, data leakage, automation mistakes, prompt-injection abuse, or a failure that causes a business, safety, or security impact. The term is useful because not every AI problem is a full breach, but many still need formal handling.
In cyber security, AI incidents matter because AI systems are now part of attack surfaces and defenses. Attackers may trigger bad outputs, manipulate a model with poisoned data, or use an AI tool to speed up phishing and reconnaissance. Defenders use incident definitions, severity ratings, and response playbooks to decide when to disable a model, notify stakeholders, preserve logs, and investigate root cause. Clear incident handling helps teams compare events consistently and respond before a small AI failure becomes a larger operational or security problem.