High-risk AI is an AI use case that can affect safety, rights, access, or major decisions, so it faces stricter legal and operational obligations. The label depends on how the system is used, not just on the model itself. A simple chatbot may be low risk, while the same model used for hiring, biometric identification, education, or access decisions can become high risk.
In cyber security, high-risk AI matters because these systems can influence authentication, surveillance, fraud detection, and privileged decisions. Attackers may target them through data poisoning, prompt manipulation, model abuse, or by exploiting weak governance around logs and human review. Defenders reduce risk by documenting use cases, limiting sensitive deployments, separating verification from identification, testing for bias and abuse, and keeping strong audit trails. Treating high-risk AI as controlled infrastructure helps organizations spot failures before they become security, compliance, or rights violations.