AI Glossary

A measure imposed by a supervisory authority to eliminate the risk or non-compliance of an AI system, for example withdrawal from the market or restriction of operation.

The most extensive part of the code, applying only to providers of GPAI models with systemic risk, which describes the process of identifying, assessing, and mitigating systemic risks before major deployment decisions.

A document that a signatory must produce before placing a model with systemic risk on the market; it contains evaluation triggers, risk categories, mitigation strategies, forecasting methods and the allocation of responsibilities.

A risk-identification method based on thinking through possible scenarios of misuse or failure of a model, supplemented by a risk inventory and consultations with internal and external experts.

An advisory body established at the EU level, composed of independent experts, that supports the AI Office and supervisory authorities in particular in the assessment of models with systemic risk.

Liability for damage arising regardless of fault, based on the mere occurrence of a harmful outcome; under consideration for the operation of risky AI systems.

A risk arising from the most capable models that may manifest at scale across the value chain, for example in cybersecurity, disinformation or misuse.