Una classe di rischio is the category used to sort an AI system by the harm it could cause in a specific use case. Under risk-based AI laws, the same model can face very different duties depending on context: a low-impact chatbot may have light transparency duties, while a system used in hiring, legal analysis, or other regulated decisions may trigger stricter controls.
This matters in cyber security because risk class drives the safeguards you must implement. Higher-risk systems usually require documentation, human oversight, logging, testing, access control, and traceability of inputs and outputs. In practice, security teams use classification to decide whether a tool can touch confidential data, whether a human must approve the result, and how much monitoring is needed to detect misuse, prompt injection, data leakage, or unreliable outputs. A correct class reduces compliance gaps; a wrong one can leave an organization exposed to legal, operational, and reputational risk.