Article 50 vs. high-risk rules — what's the difference?

These are two separate parts of the EU AI Act, and conflating them is a common source of both over- and under-worry.

Article 50 is a transparency framework: chatbot disclosure, generative-system marking, and content labelling. It's broad in reach — it applies to any provider or deployer of a covered chatbot, generative system, or published content, regardless of industry — but light-touch to comply with: a badge, a sentence, a label. It takes effect August 2, 2026, and was not delayed by the Digital Omnibus.

High-risk rules are a separate, much heavier framework that applies only to specific, defined use cases — Annex III (things like employment decisions, credit scoring, law enforcement, and similar high-stakes automated decision-making) and Annex I (AI used as a safety component in regulated products). These carry substantial obligations: conformity assessments, risk-management systems, human oversight requirements, and more. The Digital Omnibus did delay these: Annex III use cases moved to December 2, 2027, and Annex I to August 2, 2028.

There's also a narrower, separate notice duty under Article 50(3) for emotion recognition and biometric categorisation — not the same as the high-risk framework, but a genuinely different and more technical area that deserves its own legal review rather than general guidance.

For most SMEs running a support chatbot or publishing AI-assisted content, Article 50 is the relevant framework, and the high-risk rules simply don't apply — you're not doing automated employment screening or credit decisions. If you are doing something in an Annex III domain, or anything involving emotion recognition or biometrics, that's a different and heavier compliance track, and the honest advice is to get a lawyer rather than treat it as covered by transparency-duty guidance like this.

A useful mental test: Article 50 asks whether people know they're dealing with AI. The high-risk rules ask whether the AI system itself is safe and fair enough to be making, or materially influencing, a consequential decision about a person's life. Most SME chatbots and content tools answer questions and draft copy — they don't decide who gets hired, insured, or a loan — which is exactly why they land in the lighter framework.