When Words Become Obligations: Terminology, Contracts and Compliance in Artificial Intelligence

Abstract

European artificial intelligence governance is administered largely through words. Obligations attach to defined roles, evidence is assembled in documents, and claims travel through contracts and procurement files long after the conversation that produced them has ended. This research note examines a practical problem: terms that appear synonymous in ordinary professional usage — user and deployer, provider and vendor, training and AI literacy, certificate and certification, compliance and conformity — carry different legal, contractual and evidentiary weight. The method is documentary: each definition and provision is traced to a primary source, and every material claim is classified as law, official guidance, standard, accreditation practice, author analysis or author's proposed model. Two symmetrical findings organise the argument. A single substantive concept can be relabelled during the legislative process, as the 2021 Commission proposal's "user" became the adopted Regulation's "deployer". Conversely, an obligation can be materially rewritten while keeping its name and its place in the text, as Article 4 of the AI Act was by Regulation (EU) 2026/1744. A third finding runs through both: the evidentiary value of a document depends on the regime that issued it, not on the word printed on its face. The note proposes a TERM–CLAIM–EVIDENCE model — an author's operational instrument, not a legal requirement — for organisations whose vocabulary has to survive a contract, a tender file or an audit.