The EU AI Act's transparency rules, set out in Article 50, have applied since 2 August 2026. They're the part of the Act most likely to affect ordinary businesses, because they cover chatbots, AI-generated images and video, and some AI-generated text. Unlike the high-risk rules, which have been pushed back, Article 50 was not delayed.
This is a practical summary, not legal advice. If you're unsure, check the official text and the European Commission's guidance, or speak to a lawyer.
Who has to do what
Article 50 splits duties between providers (who build or supply the AI system) and deployers (businesses that use it).
| Duty | Who | What it means |
|---|---|---|
| Tell people they're talking to AI | Providers (design) and, in practice, deployers | Chatbots and voice assistants must make clear the person is dealing with AI, unless it's obvious |
| Mark synthetic content | Providers of generative AI | AI-generated audio, images, video and text must carry machine-readable marking (e.g. watermarks or metadata) |
| Disclose deepfakes | Deployers | Image, audio or video that resembles real people, places or events and could be mistaken for real must be disclosed as AI-generated |
| Disclose AI-generated public-interest text | Deployers | AI-generated text published to inform the public on matters of public interest must be disclosed, unless a human reviewed it and takes editorial responsibility |
| Inform people about emotion recognition or biometric categorisation | Deployers | People exposed to these systems must be told |
What this means for a typical small business
If you have a chatbot or AI support agent
Make it clear, at the start of the conversation, that the person is talking to AI. A simple line works: "Hi, I'm the [Company] AI assistant. I can help with orders and common questions, or connect you with a person." Also:
- Don't give the bot a human name and photo that suggest it's a person.
- Offer an easy route to a human.
- Apply the same to AI voice agents on the phone. See AI voice agents for small business.
Tools like Zelvorm Assist identify themselves as AI and hand over to your team when needed.
If you use AI images or video in marketing
- Realistic images of people, places or events that could be mistaken for real should be labelled as AI-generated, for example "AI-generated image" in the caption or on the image.
- Obviously artistic or fictional content has lighter requirements: disclosure can be done in a way that doesn't spoil the work.
- Keep the metadata that AI tools add. Don't strip watermarks or content credentials.
Read AI image generation for marketing: what's safe and legal.
If you publish AI-assisted articles
The deployer duty for text targets AI-generated text published to inform the public on matters of public interest, such as news and current affairs. It falls away where a person has reviewed or edited the content and someone takes editorial responsibility. Light proofreading or a spell-check alone is unlikely to count as meaningful review.
Good practice for any business blog: have a real expert review and edit AI drafts, and stand behind the result. That's also what Google rewards. See an AI-assisted content workflow that still sounds human.
If you build AI products
As a provider, you need to design systems that disclose they're AI and mark generated content in a machine-readable way. Systems already on the market before 2 August 2026 have until 2 December 2026 for the marking duty only.
A compliance checklist
- List every place you use AI with customers: chat, phone, email replies, images, video, articles, personalisation.
- Add clear AI disclosure to chatbots and voice agents at the start of each interaction.
- Label realistic AI images and video that could be mistaken for real.
- Keep AI watermarks and metadata intact.
- Put a human review step on any AI-generated public-facing text, and record who reviewed it.
- Update your privacy policy and terms to explain where AI is used.
- Check your vendors: ask AI providers how they handle marking and disclosure.
- Train staff on when to disclose and how.
How it fits with GDPR and other rules
Article 50 sits alongside GDPR, not instead of it. If your AI processes personal data, you still need a lawful basis, transparency, data minimisation and a data processing agreement with your provider. Read using AI tools without breaking GDPR. Consumer protection law also applies: misleading people about whether they're talking to a person, or using fake AI "customers" in reviews, can be unlawful regardless of the AI Act.
Penalties
Breaches of Article 50 can lead to fines of up to €15 million or 3% of worldwide annual turnover, whichever is higher. National authorities enforce the rules, and the Commission has published guidance on how it expects transparency to work in practice.
The short version
Since 2 August 2026, chatbots and voice assistants serving EU users must make clear they're AI, realistic deepfakes must be disclosed, and AI-generated public-interest text must be labelled unless a human takes editorial responsibility. Providers must mark synthetic content, with a grace period to 2 December 2026 for existing systems. Disclose clearly, keep humans reviewing public content and keep a record of where you use AI.
If you need a chatbot or AI feature built with disclosure designed in, our AI automation team can help, and our article on writing an AI policy for a small business gives you a starting document.




