AI Glossary

A new prohibition targeting AI systems that create or alter intimate imagery without the consent of the depicted person (so-called nudifiers), included among prohibited practices.

A new prohibition being inserted into Article 5 of the AI Act which, as of 2 December 2026, prohibits placing on the market systems designed to generate child sexual abuse material.

A measure by a supervisory authority prohibiting the further use of an AI system that violates legal requirements.

A dispute brought by authors (A. Bartz, Ch. Graeber, K. W. Johnson) against Anthropic. In June 2025, Judge William Alsup held that training a model on books is itself exceedingly transformative and constitutes fair use, but that creating and retaining a permanent library of pirated works does not.

The gradual change in the behavior of a deployed AI system during use, where it is difficult to distinguish expected adaptation from a substantial modification that would trigger a new conformity assessment.

Distortion of evaluation results caused by test tasks inadvertently ending up in the training data, so that the model achieves artificially better scores.

A systematic deviation in the data or model that may lead to unfair or discriminatory outcomes.

Assigning people to groups on the basis of biometric data; some forms (e.g. inferring sensitive characteristics) are prohibited or strictly limited by the AI Act.

The difficulty that the internal decision-making processes of complex models are very hard for humans to explain, which complicates both oversight and the proving of liability.

A party's obligation to prove the facts it asserts; in the case of harm caused by AI, its allocation is crucial, because the internal workings of the system are often difficult to prove.