Guides · Article 50 · EU AI Act
EU AI Act glossary: key Article 50 terms
Article 50 of the EU AI Act uses a handful of terms that carry a lot of weight — provider, deployer, synthetic content, deepfake, and more. This glossary explains, in plain language, how those terms are used in the transparency context, so the other guides read more easily. These are plain-language orientations, not the Act's formal definitions — several of those live elsewhere in the Regulation and can be more precise than the summaries here. This guide is informational only, not legal advice; confirm any term's exact meaning and application with qualified counsel.
Roles: provider and deployer
Article 50 splits its duties between two roles, so getting these straight matters:
- Provider — in working terms, roughly the party that develops an AI system or places it on the market. Article 50(1) and 50(2) address providers. This is a plain-language description, not the Act's full definition.
- Deployer — in working terms, roughly the party that uses an AI system under its own authority. Article 50(3) and 50(4) address deployers. Again, a working description rather than the formal definition.
- A single business can be a provider of one system and a deployer of another, so the role is assessed per system.
System and content terms
The terms that describe what a system does or produces:
- AI system — the software system the obligations attach to; the Act's formal definition sits elsewhere in the Regulation and governs the precise boundary.
- Synthetic content — audio, image, video, or text generated by an AI system. Article 50(2) asks providers to mark such outputs in a machine-readable format, detectable as artificially generated or manipulated.
- Deepfake — in Article 50(4), image, audio, or video content, generated or manipulated by an AI system, that constitutes a deep fake; deployers must disclose it as artificially generated or manipulated. Whether a given output is a 'deep fake' is fact-specific.
Biometric and emotion terms
Two terms appear together in Article 50(3), both tied to existing EU data-protection law:
- Emotion recognition (system) — a system that infers or identifies emotions or intentions of natural persons. Article 50(3) asks deployers to inform the persons exposed and to process personal data under the referenced instruments.
- Biometric categorisation (system) — a system that assigns natural persons to categories on the basis of biometric data. It sits alongside emotion recognition in Article 50(3)'s deployer duty.
- For both, whether a given tool meets the Act's definition is a fact-specific threshold question, and the formal definitions live elsewhere in the Regulation.
Common questions
Are these the official EU AI Act definitions?
No. This glossary gives plain-language orientations for how the terms are used in Article 50's transparency context. The Act's formal definitions live elsewhere in the Regulation and can be more precise. Confirm any term's exact meaning and how it applies to your situation with qualified counsel.
Where's the line between provider and deployer?
As a working distinction, a provider roughly develops or places an AI system on the market, and a deployer roughly uses one under its own authority — but these are plain-language descriptions, not the full legal definitions, and the role is assessed per system. Confirm your role with qualified counsel.
See what may apply to your business
Answer seven quick questions for an automated, informational indication of which Article 50 obligations appear likely to apply — free, and not legal advice.