AI Glossary

A European directive laying down cybersecurity requirements for selected organizations.

An organization subject to the obligations under the NIS2 Directive.

A voluntary framework of the U.S. National Institute of Standards and Technology for managing risks associated with AI. It is sector-neutral, adaptable to the size of the organization, and serves as an operational layer beneath binding regulation rather than as a substitute for it.

A supplementary part of the NIST framework devoted to the specific risks and measures associated with generative AI, which it maps to the individual functions of the framework.

An interference with a person's rights or personal sphere that cannot be directly quantified in monetary terms.

A tool by Google that allows users to analyze their own documents and generate answers, summaries, and research over them.

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.

An authority that assesses, designates, and oversees conformity assessment bodies (notified bodies) in the field of AI.

A conceptual division of an AI system's existence into phases, from design and data collection through development and deployment to operation and decommissioning, used to assign risks and responsibilities.

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

A tool that allows AI systems to be described and compared along dimensions such as data, the role of humans, the deployment context, and potential impact on society.

Running a model on an organization's own infrastructure instead of the public cloud, chosen in particular for reasons of data confidentiality and control.

A model whose weights or code are publicly available, allowing it to be self-deployed, modified, and fine-tuned.

A license allowing software to be freely used, distributed and modified under specified conditions, also applied to some AI models and tools.

A model whose trained weights are publicly available for download and self-deployment, even though its training data or code need not be published.

OpenAI's programming interface, which allows developers to integrate its models into their own applications and services.

The ability of a rightholder to reserve that its works may not be used for text and data mining; the reservation must be expressed in a machine-readable manner.

A situation in which a model adapts too closely to the training data and its ability to work with new data deteriorates.

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.

The number of a model's internal variables (weights); it tends to be an indicator of its size and capacity.

Legal protection of a technical solution that meets the statutory conditions.

Harm expressible in monetary terms suffered by the injured party as a result of a certain event, for example the failure of an AI system.

Simulated attacks used to verify the security of a system.

An AI search engine combining language models with internet search and the citing of information sources.

Any information relating to an identified or identifiable natural person; its processing in the development and operation of AI is subject to data protection rules.

An entity that processes personal data on behalf of the controller.

Personality rights protecting a person's identity, likeness, privacy, and dignity.

Protected works obtained from illegal sources and used for training; the mere downloading and retention of such copies has been assessed by courts more strictly than the training itself.

The first making available of an AI system on the Union market, whether for payment or free of charge.

The unauthorized appropriation of the authorship of another person's work or a substantial part of it.

We explore where the line between inspiration and plagiarism lies in our article Robbing a work is not inspiration!