AI Glossary

A term for the legal questions surrounding the copyright protection of prompts.

Methods of creating and optimizing instructions that lead to higher-quality AI outputs.

A security attack in which an attacker inserts hidden instructions into the input in order to bypass the model's intended behavior or its restrictions.

The contested question of whether and when the outputs of a model themselves can enjoy copyright protection, and to whom any rights belong.

We examine how courts have approached the question in our articles A Czech court has ruled that AI-generated graphics are not a copyrighted work and Thaler v. Perlmutter.

Indirect discrimination in which a model uses a seemingly neutral data point that is nonetheless closely related to a protected characteristic (e.g. a ZIP code as a substitute for ethnicity).

The modification of data so that a specific individual cannot be identified without additional information.

Data freely accessible on the internet whose use for AI training may nevertheless be subject to copyright or personal data protection rules.

A principle under which personal data may be processed only for compatible purposes; using previously collected data to train AI raises the question of compatibility.

The commencement of use of an AI system for its intended purpose; the moment from which the relevant obligations apply to its operation.