AI Glossary

A concept of AI that is lawful, ethical, and technically robust; it includes, among other things, respect for fundamental rights, transparency, fairness, and human oversight.

An environment in which rules, oversight and user trust support the uptake of responsibly developed and deployed AI; a goal toward which European regulation is directed.

An intergovernmental set of value-based principles for trustworthy AI that has become a reference point for many national and supranational frameworks and unifies the basic terminology.

A method of calculating a penalty in which the maximum fine is based on a company's total worldwide annual turnover, so that the sanction remains effective even against large companies.

Fragmented regulation at the level of individual U.S. states (e.g., California, Colorado, Illinois, Texas), which, alongside the federal level, introduces its own obligations for certain uses of AI.

An advisory body established at the level of the United Nations that addresses the global governance of artificial intelligence and recommendations for international cooperation.

The highest risk category of the AI Act, comprising practices that are entirely prohibited because they conflict with EU values and fundamental rights.

An international document setting out ethical principles for the development and use of AI, emphasizing human rights, human dignity, and the protection of the environment.

Machine learning that searches for patterns in unlabeled data without having correct answers assigned to it in advance.

A prohibited practice consisting of the mass and untargeted scraping of facial images from the internet or CCTV footage in order to create facial recognition databases.

A presidential act in the US setting the direction of federal policy in the field of AI; its content changes according to the priorities of individual administrations and complements sectoral regulation.

Data used continuously during the development of a model to tune its settings and to continuously verify its performance.

An effort to ensure that a model's goals and behavior align with human values and intentions, thereby preventing undesirable or harmful conduct.

A database optimized for storing and searching vector representations (embeddings), which is the basis of vector and semantic search.

Search based on the similarity of semantic representations (embeddings) rather than keyword matching; the basis of modern RAG systems.

The management of risks associated with vendors of AI solutions and services, including the assessment of their reliability, security, and legal compliance.

Models for creating video from a text prompt; Sora comes from OpenAI, Veo from Google. They are associated with licensing agreements as well as concerns about deepfakes.

An artificially created, realistic digital character capable of interaction that appears to be a real person.

An artificially created digital character that appears on social media as an influencer and has no real physical counterpart.

A technology for the artificial imitation of a specific person's voice; related to the protection of personality rights, fraud risks, and deepfakes.

Rights protecting a person's voice against unauthorized imitation or commercial use.

Protection against the unauthorized imitation of a specific person's voice; related to personality rights as well as the risks of fraud and deepfakes.

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

A supplier's contractual assurance that the data used for training was obtained and used in accordance with the law, often supplemented by an indemnification clause.

The embedding of a hidden or machine-readable marker into generated content that makes it possible to recognize that it was created by artificial intelligence.

The automated retrieval of content from websites, used among other things to collect data for training models, which raises legal questions.

A model for automatic speech transcription and translation, used to convert spoken words into text.

The reporting of violations of AI-related rules; the protected status of whistleblowers is one of the tools for overseeing compliance with the AI Act.

A measure aimed at removing an AI system from the distribution chain.

An agent that manages and automates a sequence of steps in a workflow and may coordinate other tools or agents.