AI Glossary

The distinction between evaluating what a model can do (its capabilities) and evaluating whether its use is safe; both form part of managing systemic risks.

Concerns that advanced models may facilitate access to information usable for chemical, biological, radiological and nuclear threats; one of the categories of systemic risk.

A marking indicating a product's conformity with EU requirements; high-risk AI systems must bear it after a successful conformity assessment.

Technical standards being prepared by the European standardization organizations, compliance with which establishes a presumption of conformity of high-risk systems with the requirements of the AI Act.

System design in which legal and security requirements, defined purposes of use, risk management and ongoing auditability are built in from the development stage onwards, so that the agent remains compliant with regulation.

A structured set of controls from which an organization selects those that match the risks of a particular AI system and the requirements of several frameworks at once.

A step taken to eliminate an identified non-compliance or risk associated with an AI system.