AI Glossary

A technical standard adopted by the European standardization organizations, compliance with which establishes a presumption of conformity of a high-risk AI system with the requirements of the AI Act.

An AI system falling into the high-risk category under the AI Act, subject to strict requirements on risk management, data quality, documentation, transparency and human oversight.

The obligation to report serious incidents and malfunctions of an AI system to the competent authorities within set deadlines, so that risks can be assessed and mitigated.

Automatically created records of an AI system's activity that enable retrospective checking, auditing and incident handling; they are mandatory for high-risk systems.

A principle under which a human must have real, not merely formal, control over a high-risk system, including the ability to intervene in and override the system.

An independent body tasked with assessing the conformity of high-risk AI systems with the requirements of the AI Act before they are placed on the market.

The risk that an advanced system trained to optimize a certain goal begins to circumvent control mechanisms or human oversight in order to achieve the goal; particularly monitored in agentic systems.

A documented procedure by which the provider systematically monitors the functioning of a high-risk system after deployment and evaluates newly identified risks.

A documented procedure by which the provider systematically monitors the functioning of a high-risk system after deployment and evaluates newly identified risks.

A process in which an already supplied AI system is withdrawn from users due to identified risks or non-conformity.

The obligation to enter a high-risk AI system into a public EU database before it is placed on the market or put into service, serving transparency and oversight.

A set of processes through which a provider continuously identifies, assesses, monitors and mitigates the risks associated with an AI system.

The core regulatory logic of the AI Act, which classifies systems according to their level of risk into categories ranging from unacceptable risk (prohibition) through high risk (strict obligations) and limited risk (transparency obligations) to minimal risk.

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.

An example of US state legislation targeting high-risk systems and algorithmic discrimination, imposing duties of care and duties to inform affected individuals on deployers.

The supervisory authority competent for the use of high-risk AI systems in the financial sector.

A package of targeted amendments to the AI Act agreed in 2026, which in particular postpones the applicability of the obligations for high-risk systems and introduces new prohibitions. The aim is simplification and more time to prepare standards.

A person whose age, health condition, or social situation increases the risk of being negatively affected by an AI system.