Does your business run a chatbot on its website, generate visuals with AI, or write copy with ChatGPT? Since August 2, 2026, Article 50 of the European AI Act requires you to say so. This is not an option or a legal footnote: it is a transparency obligation that applies to any company, with no headcount or revenue threshold. Here is what the rule says, in practice, and how an SME gets compliant.
In brief
- Article 50 of the AI Act has been fully applicable since August 2, 2026: it requires disclosing chatbots, AI-generated or manipulated content, and deepfakes.
- It applies to all companies, regardless of size, as soon as a customer interacts with AI or a company publishes AI-generated content.
- A technical grace period runs until December 2, 2026 for machine-readable marking of systems already on the market before August 2026.
- Penalties reach 15 million euros or 3% of worldwide turnover, with the lower amount applied to SMEs.
- Compliance comes down to four simple actions, achievable in-house without a specialist lawyer.
Article 50, in one sentence
Article 50 of the AI Act is the provision of the European AI regulation that requires transparency whenever a person interacts with artificial intelligence or consumes content it produced. Unlike the rules on high-risk systems, it does not only apply to sensitive sectors: it covers any everyday use of a chatbot, an image generator, or a writing assistant, regardless of the company's line of business.
August 2, 2025
Groundwork
August 2, 2026
Article 50 applies
December 2, 2026
Technical grace period ends
Four concrete obligations
According to the European Commission, Article 50 breaks down into four distinct requirements. Each addresses a specific use case.
| Obligation | Who is concerned | Concrete example |
|---|---|---|
| Disclose an AI interaction | Providers of chatbots, voice assistants, conversational agents | A customer-support chatbot must state from its first message that it is an AI |
| Technically mark generated content | Providers of generative AI (text, image, audio, video) | A visual created with a generative AI tool carries a watermark or detectable metadata |
| Disclose emotion recognition or biometric use | Deployers of these systems | Software analyzing a customer's tone of voice in a call center must announce it |
| Label deepfakes and AI text on public-interest topics | Deployers of AI content on sensitive topics | An AI-generated article on health or politics must carry a visible label, unless reviewed by an editor who takes responsibility for it |
The most common obligation for an SME remains the first one: if your website offers a chatbot or voice assistant, the person interacting with it must understand, from the start, that they are not talking to a human. The wording can be simple: a welcome message is enough, as long as it isn't buried in unreadable terms of service.
Who is concerned: no size threshold
Unlike other parts of the AI Act, Article 50 sets no headcount or revenue threshold. A ten-employee SME is just as concerned as a large group, as soon as it runs a chatbot, generates marketing images with AI, or publishes text written by a language model. The distinction between provider (who develops the tool) and deployer (who uses it professionally) still matters: most SMEs are deployers, but the disclosure obligation still falls on them directly in their relationship with their own customers.
Audit your AI touchpoints
Add a clear disclosure
Turn on technical marking
Train and document
What non-compliance costs
As with the rest of the AI Act, the proportionality principle set out in Article 99 applies: between the 15 million euros and the 3% of worldwide turnover, the lower amount is applied for an SME. This remains a deterrent, but the Commission's goal is not to primarily target small, good-faith organizations: national market surveillance authorities focus first on repeated or deliberate breaches.
Good news: compliance stays largely achievable without a dedicated lawyer. Most consumer-facing tools (ChatGPT, Copilot, Mistral AI) already include marking features or easy-to-enable disclosure settings. The main effort for an SME is organizational: identify its uses, add the disclosures, and document the process.
FAQ
Is a ten-employee SME really concerned by Article 50?
Yes. Article 50 of the AI Act sets no headcount or revenue threshold. As soon as a company has a customer interact with a chatbot or publishes AI-generated content, the transparency obligation applies, regardless of its size.
Does a text written with the help of ChatGPT need to be disclosed?
It depends on the use. A public-interest text (health, politics, justice, environment) generated by AI must be disclosed, unless it went through editorial review and validation by a person who takes responsibility for it. A simple internal draft reworked by a human is not concerned.
What happens to content generated before August 2, 2026?
According to the European Commission, content produced before that date does not need to be labeled retroactively. Only new content and interactions, from August 2, 2026 onward, are covered by the obligation.
How does this relate to the December 2, 2026 deadline?
December 2, 2026 only concerns the machine-readable technical marking (watermarks, metadata) of generative AI systems already on the market before August 2026. The human-facing disclosure obligation (chatbot, deepfake) has applied since August 2, 2026, with no additional delay.
Summary
Article 50 of the AI Act formalizes a common-sense rule: tell your customers when they are talking to a machine or reading AI-generated content. For an SME, compliance comes down to a few concrete actions, without a major technical overhaul. It also has a business upside: a company transparent about its AI use inspires more trust than one that hides it.
To go further on AI Act deadlines and AI compliance for SMEs, check out our other resources on AI for business.
Sources: European Commission (digital-strategy.ec.europa.eu, FAQ on transparency obligations under Article 50 of the AI Act); Addleshaw Goddard, "AI transparency under the AI Act: what businesses need to know before 2 August 2026."
