AI Glossary

A reservation of rights under the rules on text and data mining, by which a rightsholder prevents the use of its content for automated analysis and training. Want to learn more? Our colleague Hana Krajčíková has prepared, for Section 39c of the Copyright Act, a commentary on this issue on Zákony pro lidi.

A set of documents describing the design, operation, purpose and legal compliance of an AI system.

A set of technical rules used to demonstrate compliance with legal requirements.

A parameter influencing the randomness of the output; a lower value leads to more predictable responses, a higher value to more varied and creative ones.

Data used to verify whether an AI system achieves the required results after the completion of development.

Computing power expended at the response stage (not during training), the increase of which can improve output quality for demanding tasks.

The automated analysis of texts, images or data in order to extract information. Want to know more? Our colleague Hana Krajčíková has written a commentary on this topic on Section 39c of the Czech Copyright Act for Zákony pro lidi.

A provision of the EU Directive on Copyright in the Digital Single Market that, under certain conditions, permits the automated analysis of protected works; it may be used for commercial purposes provided that the rights holder has not reserved such use.

A US case addressing whether a work created entirely by a machine, with no human author, can enjoy copyright protection. The courts upheld the US Copyright Office's position that human authorship is a prerequisite for protection.

We analyzed the case in detail in our article Artificial intelligence is not the author of a work – Thaler v. Perlmutter.

The phenomenon whereby European AI rules effectively become a global standard, because multinational providers apply them outside the EU as well.

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 adaptation act prepared by the Czech Ministry of Industry and Trade implementing the AI Act into Czech law. It is minimalist in nature, imposes no new technical requirements and mainly governs institutional, procedural and enforcement mechanisms.

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

The authority designated in the Czech draft as the principal supervisory authority and single point of contact for the AI Act, coordinating oversight of compliance with the rules.

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 U.S. legal institution that, under certain conditions, permits the use of a protected work without permission. In AI disputes, courts assess in particular the transformativeness of the use and its market impact; the outcome strongly depends on how the data was obtained and what it is used for.

A lawsuit filed by The New York Times in December 2023 in a federal court in Manhattan, in which the newspaper claims that OpenAI and Microsoft unlawfully used millions of its articles to train the ChatGPT and Copilot models and that the systems reproduce protected content verbatim. OpenAI, by contrast, argues that the plaintiff deliberately steered the models with deceptive prompts in order to elicit the reproduction of text.

The internationally recognized definition of an AI system on which a number of legal frameworks, including the European one, are based; it emphasizes the machine-based generation of outputs influencing environments with varying degrees of autonomy.

The supervisory authority for personal data protection, which within AI oversight is to be responsible in particular for matters related to the processing of personal data and biometrics.

A continuous improvement methodology underpinning management system standards; it consists of planning, implementation, verification of results and subsequent correction, in a repeating cycle.

Verification of an AI system's compliance by an independent body.

A dispute in which Thomson Reuters accused the Ross Intelligence service of unlawfully using so-called headnotes from the Westlaw database to train a competing legal search engine. In February 2025, the court ruled that the copying was not fair use, among other reasons because Ross was not a generative AI and directly competed with Westlaw.

The basic unit of text that a language model works with when processing and generating content.

The process of splitting text into smaller units (tokens) that the model processes; a token can be a word, part of a word, or a character.

An agent that uses external tools, interfaces (APIs), or functions to accomplish tasks.

The ability of an AI model to use external applications, databases, or services while performing a task.

A method of selecting the next token from the smallest set of options whose combined probability reaches a specified value P.

The ability to retrospectively document what data and what process an AI system used, through records of data sources, their modifications, and metadata; it supports analysis and accountability.

Confidential business information having actual or potential value, which its owner deliberately protects against disclosure.

A sign distinguishing the goods or services of one business from those of another, protected by industrial property law.