High-risk AI covers AI systems that the EU AI Act considers capable of significantly affecting people's health, safety or fundamental rights. They are permitted, but subject to requirements for documentation, data quality and human oversight.
The regulation lists a number of areas where AI is high-risk by default. They include recruitment and personnel management, education and examinations, access to public services and credit, and law enforcement. What matters is what the system is used for. The same language model can be harmless in a newsletter and high-risk when it sorts job applicants.
Most requirements fall on the provider, who must document the system. Organisations that deploy a high-risk system also have obligations: they must use it according to the instructions, ensure human oversight, monitor how it works, and inform employees before the system is put into use in the workplace. Following the AI Omnibus, the requirements for areas such as employment and education apply from 2 December 2027.
In practice
For most communication and leadership tasks, high-risk is not relevant. It becomes relevant when AI starts to feed into decisions about people: who is invited to interview, who receives a service, and how an employee's performance is assessed. Ask early whether a new tool makes or prepares decisions of that kind.
The risk categories are explained in my guide to the EU AI Act.
The concepts in this glossary run through my AI courses and workshops, where we work with them on your own tasks.