Compliance · 8 October 2026
EU AI Act Article 50: what small businesses must do
If you use a chatbot or AI agent that talks to people, one short article of the EU AI Act probably applies to you. It has applied since 2 August 2026, and the delays you heard about did not cover it.
What Article 50 requires
Article 50 of Regulation (EU) 2024/1689 sets transparency duties that sit outside the Act's risk tiers. For chatbots and similar systems that interact directly with people (Article 50(1)), the person must be informed they are interacting with an AI system, unless that is obvious from the context. Other limbs cover marking of synthetic content, emotion-recognition notices and deepfake labelling.
The dates
- 2 August 2026: Article 50 transparency duties apply and national authorities can enforce them.
- 2 December 2027: Annex III high-risk obligations, deferred by the Digital Omnibus.
- 2 August 2028: high-risk systems embedded in regulated products (Annex I).
The Omnibus left chatbot disclosure on schedule. Reporting suggests separate transitional relief for machine-readable marking of synthetic content for systems already on the market.
Does it apply to a UK business?
If your AI talks to people in the EU, assume yes. If you serve only UK customers, the Act does not apply to that activity, and UK GDPR remains your main obligation. Telling customers they are talking to an AI is good practice either way.
Provider or deployer?
The vendor that builds the system is the provider. The business that uses it with its own customers is the deployer. A small business using a vendor's agent is normally the deployer, so check what the vendor discloses and do not switch it off or disguise it.
A practical checklist
- Disclose in the first message. For example: “Hi, I'm Leo, the AI assistant for [Business].”
- Disclose in each channel. WhatsApp profile text, website chat banner, email signature, and the opening of any voice interaction.
- Offer a person. Make the route to a human clear for complaints and anything sensitive.
- Do not impersonate a human. No human names, photos or claims for an AI agent.
- Flag staff-facing use. Tell employees when they interact with an AI system.
- Be careful with decisions about people. Using AI to make or materially influence hiring, firing or credit decisions can move you into higher-risk territory.
- Train your team. The AI literacy duty has applied since February 2025.
- Keep records. Note what you disclose and where.
Penalties
Transparency breaches can be fined up to €15 million or 3% of worldwide annual turnover, whichever is higher; for SMEs the lower figure applies. National authorities enforce.
How Aiconomy handles it
Every customer-facing message from Leo, Ava or another agent identifies the sender as an AI employee by name and role, live chat and voice interactions open with a clear AI statement, and staff-facing use is disclosed in onboarding. See the full regional compliance position, including open items. Agents start supervised, so a human approves consequential actions.
Questions
Does Article 50 apply to a UK business?
It can. The EU AI Act applies where AI systems are used in the EU, so a UK business whose chatbot or AI agent talks to people in the EU should assume Article 50 applies. As far as we know, the UK has no equivalent statute at the time of writing, but UK GDPR still applies to the personal data involved. Check the current position.
What counts as clearly telling someone they are talking to an AI?
Say so at the start of the conversation, in plain words, in the same channel. The Act allows an exception where it is obvious from the circumstances, but it is safer not to rely on that.
What is the penalty?
For transparency breaches, fines of up to €15 million or 3% of worldwide annual turnover, whichever is higher. For SMEs, the Act provides that the lower of the two figures applies. Enforcement is by national authorities.
Did the Digital Omnibus delay Article 50?
Not the chatbot disclosure duty. The Omnibus deferred high-risk obligations (Annex III to 2 December 2027, Annex I products to 2 August 2028). Reporting indicates separate transitional relief for the machine-readable marking duty in Article 50(2) for systems already on the market, not for chatbot disclosure.