AI-made content: what the AI Act requires you to disclose since 2 August 2026

This article was translated from Portuguese. For legal references, the Portuguese version prevails.
In short
- The obligations are in Article 50 of Regulation (EU) 2024/1689 and have applied since 2 August 2026.
- The Digital Omnibus on AI (Regulation (EU) 2026/1744) did not postpone transparency. It only gave providers of generative tools already on the market until 2 December 2026 to mark their content.
- Chatbot on your website or on WhatsApp: it must be clear that it is AI, unless that is obvious.
- Deepfakes (people, places or events that look real): whoever publishes them must say they were generated or manipulated.
What started to apply on 2 August 2026?
Regulation (EU) 2024/1689, known as the AI Act, applies in stages. 2 August 2026 was the general date of application, and with it came Article 50, on transparency. Unlike the rules for high-risk systems, this article applies to any AI system that interacts with people or generates content, whatever its risk.
In July, the Union adopted the so-called Digital Omnibus on AI, Regulation (EU) 2026/1744, published on 24 July and in force since 27 July 2026. It postponed the obligations for high-risk systems to 2 December 2027 and 2 August 2028, but kept the transparency date. The only leeway is for providers of generative systems placed on the market before 2 August: they have until 2 December 2026 to mark the content they produce.
What are the obligations in Article 50?
The article distinguishes the provider (whoever develops the system, or places it on the market under their own name) from the deployer (whoever uses it in their business). An SME is almost always the second, but it can be the first: if it builds an assistant on a third-party model and runs it under the company’s name, it may be considered the provider of that assistant.
| Situation | Who has the obligation | What they must do |
|---|---|---|
| System that interacts directly with people (chatbot, voice assistant) | Provider | Design it so that people know they are talking to an AI, unless it is obvious from the context |
| System that generates text, images, audio or video | Provider | Mark the content in a machine-readable way, so it can be detected as AI-generated |
| Emotion recognition or biometric categorisation | Deployer | Inform the people exposed to it and comply with the GDPR |
| Deepfake: image, audio or video that imitates real people, objects, places or events and appears authentic | Deployer | Disclose that the content has been artificially generated or manipulated |
| AI-generated text published to inform the public on matters of public interest | Deployer | Disclose that it was AI-generated, unless it has undergone human review and someone holds editorial responsibility |
The information must be given in a clear and distinguishable way, at the latest at the first interaction or exposure, and must meet accessibility requirements. For evidently artistic, satirical or fictional works, the deepfake obligation is limited to disclosing the existence of the generated content, in a way that does not spoil the work.
In practice: what changes for an SME?
Chatbot on your website, WhatsApp or Messenger
This is the most common case. The person must know they are talking to a machine. The simplest way is to say so in the first message: “I am company X’s automated assistant”. A notice hidden in the terms is not enough. And if the assistant presents itself with a person’s name and photo, the risk of misleading people increases.
Product images and ad creatives
Technical marking (the digital watermark or the metadata) is the obligation of whoever provides the tool. On your side, the question is a different one: does the image show real people, places or events, and could it pass for genuine? If so, it is a deepfake and must be identified. A generated studio background for photographing a real product is unlikely to mislead anyone. A “satisfied customer” who never existed, or your shop with a window display it does not have, is another matter, and also falls under the rules on unfair commercial practices.
Voices and videos with synthetic people
A video with a realistic avatar or a cloned voice of a real person is the typical example of a deepfake. Besides saying it was AI-generated, a real voice or face requires that person’s consent. Our own podcast has drawn characters and AI-created voices, and each episode says “Video created with AI” on screen.
Blog and social media texts
A commercial text written with the help of AI and reviewed by someone at the company has no labelling obligation under Article 50. The text rule applies to publications that inform the public on matters of public interest, and even there it falls away if there is human review and editorial responsibility. But review is what protects the business: a wrong fact, once published, belongs to the company, not to the tool.
How much can non-compliance cost?
For these obligations, Article 99 provides for fines of up to 15 million euros or 3% of total worldwide annual turnover, whichever is higher. For SMEs, the lower of the two applies. More than the fine, the real risk for a small business is trust: a customer who finds out that the “testimonial” was synthetic does not come back.
What to do this week
- List the places where you use customer-facing AI: chatbot, automatic replies, images, videos, voices.
- For each assistant, check that the first message says it is automated.
- Review the images and videos you have published: do any show real people, places or situations that did not happen? Identify them or take them down.
- Write a short internal rule: what may be generated, what must be identified and who reviews it before publication.
- Ask the providers of your tools to confirm how they mark generated content.
- Consult the Commission’s guidelines on the transparency obligations, from July 2026, for borderline cases.
AI is still a good help for producing more and better. The rule is simple to follow when you think about it from the start: tell people when it is a machine, and never pass off as real what is not.
Frequently asked questions
Has the AI Act been postponed?
In part. The Digital Omnibus on AI (Regulation (EU) 2026/1744) postponed the rules for high-risk systems to 2027 and 2028. The transparency obligations of Article 50 have applied since 2 August 2026; only the marking of content by generative tools already on the market has a deadline of 2 December 2026.
Do I have to put “AI-generated” on every image I use?
No. Technical marking is the obligation of the tool’s provider. Whoever publishes only has to identify deepfakes: images, audio or video that imitate real people, objects, places or events and could pass for authentic.
Does my chatbot have to say it is an AI?
Yes, unless it is obvious from the context. The safest option is to say so clearly in the first message.
Do blog texts written with AI have to be identified?
Under Article 50, only texts published to inform the public on matters of public interest, and even those need no label if they have human review and editorial responsibility.
Sources
- Regulation (EU) 2024/1689 (Artificial Intelligence Act), EUR-Lex
- Regulation (EU) 2026/1744 (Digital Omnibus on AI), EUR-Lex
- European Commission: guidelines on the transparency obligations (July 2026)
- Goodwin: “EU AI Act Transparency Obligations Are Now in Force” (August 2026)
This article is general information, updated on 18 November 2026, and does not replace legal advice on your case.
Written and reviewed by the VisãoValor team, with the sources listed at the end.


