Guides · Article 50 · EU AI Act
AI content moderation and EU AI Act Article 50
AI content moderation systems scan posts, images, or messages and flag or remove those that break the rules. Teams ask whether Article 50 of the EU AI Act requires disclosing that moderation is AI-driven. On a typical description, a classifier that labels content may not fit any of Article 50's four situations — a fact-specific reading, and other parts of the Act could still apply. This guide explains the reasoning. It is informational only, not legal advice.
Test moderation against the four situations
Article 50's transparency duties attach to systems that interact directly with people (50(1), providers), generate synthetic content (50(2), providers), perform emotion recognition or biometric categorisation (50(3), deployers), or produce deepfakes and certain public-interest text (50(4), deployers). A moderation system that classifies whether content violates a policy generally does none of these: it does not hold itself out to a user as a conversational AI, and it does not generate synthetic audio, image, video, or text. On that description it may fall outside Article 50.
Watch the edges — and the rest of the Act
Two cautions apply. First, if a moderation product also generates content — for example an AI-written explanation sent to a user, or a chat appeal handler — those elements could independently engage 50(2) or 50(1). And if a system categorises people using biometric data, Article 50(3)'s deployer-side duty could come into view; whether a given moderation tool does that is a fact-specific question about what it actually processes. Second, Article 50 is only the transparency slice of the EU AI Act. Other parts of the Regulation — outside Article 50 and outside this guide — may still apply to a moderation system. So the accurate framing is 'may fall outside Article 50's transparency obligations,' not 'the Act does not apply.'
How to document the position
Describe what your moderation system does — classify, flag, remove, explain — and what, if anything, it generates or how it categorises people. Map each element to the four situations, and where nothing fits, record that reasoning rather than assuming. Confirm with qualified counsel; our free scope check helps frame the analysis but does not decide whether the system is in or out of scope.
Common questions
Do we have to tell users that moderation is done by AI?
A classifier that only flags policy-violating content may fall outside Article 50's four situations, though that is fact-specific and other parts of the EU AI Act could apply. Generative or biometric-categorisation elements could change the analysis. Confirm 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.