Regulation (EU) 2024/1689, Article 50
For marketing, content and customer service teams using AI, and working out which content and chatbots need disclosing. This page covers what you need to know and ways to avoid running into issues.
What we've seen reported
Article 50 of the EU AI Act, Regulation (EU) 2024/1689, sets transparency obligations for chatbots and AI-generated content, from telling people they're interacting with an AI to disclosing deepfakes.1 The European Commission published guidelines on it in July 2026.2
Last updated October 2026 · General information only, not legal advice. Sources.
What the rules cover
These are the kinds of content Article 50 of the EU AI Act describes, according to its text, and the disclosure each is described as needing.1
| Content | Label it needs |
|---|---|
| Chatbots and AI assistants that interact with people | Providers: people are told they're interacting with an AI, unless that's obvious from the context. |
| Synthetic audio, images, video and text from generative AI systems | Providers: outputs are marked in a machine-readable format and detectable as AI-generated. |
| Images, audio and video that are deepfakes | Deployers: disclosed as artificially generated or manipulated. |
| Deepfakes in evidently artistic, creative, satirical or fictional work | Deployers: a lighter disclosure that doesn't hamper the display or enjoyment of the work. |
| AI-generated text published to inform the public on matters of public interest | Deployers: disclosed as AI-generated, unless it has had human review or editorial control and someone holds editorial responsibility. |
| Emotion recognition and biometric categorization systems | Deployers: people exposed to them are informed that the system is operating. |
| Every disclosure above | Given in a clear and distinguishable way, at the latest at the first interaction or exposure. |
Where it goes wrong
One approved label ends up in many channels, markets and languages. Along the way it gets shortened, translated and reworded, and some versions end up losing the disclosure these rules require.
Spring campaign image, labeled "Created with AI"
How Arbitr helps
Manually review everything? Or do what modern teams do and use AI to check that what you publish matches what you approved.
FAQ
In more places than most teams track. Article 50 of the EU AI Act covers chatbots, deepfakes and some AI-written text,1 and we usually find AI-made content in:
Often they don't. A label added to the master image disappears when someone crops it for a square social format, re-edits the video, or reposts it on a marketplace. Labels built into the content itself tend to last longer than ones added around it.
More on checking versions at scale is on our content checks page.
More than the team that bought the AI tool. Content, social, e-commerce and customer service (for chatbots) all publish what Article 50 describes, and so do your agencies. A short briefing for each, using examples from your own content, does more than a policy document. Related rules are on our AI content disclosure hub.
Keep a simple record for each asset: which tool made or edited it, who approved it, which label it carries and where it's published. When a version turns up without its label, that record lets you trace where it came from.
How it applies to a specific asset is a question for your legal adviser. Where that asset appears across your markets and languages is a question Arbitr answers.
Mostly not, despite the headlines. The Digital Omnibus on AI, Regulation (EU) 2026/1744, moved most high-risk rules to December 2, 2027, and left Article 50 applying from August 2, 2026.4
Its one change for Article 50 is a new Article 111(4): generative AI systems already on the market have until December 2, 2026 for machine-readable marking.4
They're separate parts of the same regulation. The European Commission describes the AI Act as sorting AI systems into risk categories, with systems that pose transparency risks covered by Article 50, apart from the requirements for high-risk systems.2
Arbitr checks your content against the rules, claims and terms your team has approved, and reports where something doesn't match, by channel, market and language. It reports matches against your rules, not legal verdicts. What to do with each one stays with your team.
