Knowledge base
Labelling AI content: what Article 50 of the AI Act asks of an SME
By Bob Biesemans · published on
Since 2 August 2026 the AI Act's transparency rules apply. When must an SME say that an image, video or text was made with AI, and when not?
Short answer
Since 2 August 2026, anyone who uses AI to make a realistic image, audio or video of real people, places or events (a deepfake) must disclose that it is artificial. AI texts published to inform the public on matters of public interest also get a disclosure, unless a person reviewed them and someone holds editorial responsibility.
In force since 2 August 2026
Article 50 of the AI Act contains transparency rules. The Digital Omnibus did not postpone them: they apply since 2 August 2026. Some are aimed at the makers of AI systems, some at organisations that use AI, such as an SME.
What applies to an SME that uses AI
Deepfakes: anyone who uses AI to create or alter images, audio or video that resemble real people, places or events and could seem authentic must disclose that it is artificial. For clearly artistic, creative or satirical work, a disclosure that doesn't spoil the work is enough.
Texts: anyone who uses AI to create or alter text published to inform the public on matters of public interest must disclose it, unless a person reviewed the text and someone holds editorial responsibility.
What usually falls outside the duty
A quote, an internal memo or a customer email you drafted with AI and checked yourself usually falls outside this labelling duty. So does, usually, a product photo touched up with AI without showing anything misleading. Honesty with customers is still wise, and misleading advertising remains prohibited.
Make it a rule
Put in your AI policy who decides whether something gets a disclosure and what it looks like. Train the people who publish, such as marketing and sales, to recognise realistic AI images and public-facing texts. Measure whether they apply it in practice. General information, not legal advice.
Frequently asked questions
Do I have to say a customer email was written with AI?
Usually not under Article 50: it covers deepfakes and texts published to inform the public. An email you checked yourself generally falls outside it.
Does the labelling duty apply to advertising images?
If an AI image imitates real people, places or events and could seem authentic, yes. For clearly creative work a disclosure that doesn't spoil the work is enough.
Did the Digital Omnibus postpone Article 50?
No. The transparency rules apply since 2 August 2026. Only certain providers of AI systems get a transition for part of the rules.