The EU Artificial Intelligence Act (Regulation (EU) 2024/1689) has a short list of transparency rules — Article 50 — that apply from 2 August 2026: people must be told when they are talking to an AI system, and AI-generated content must be marked so software can recognise it. If you use Vex, an AI phone agent, AI email answering or AI replies on Facebook and Instagram, these rules apply to your business. The good news: the disclosures are built into Yaplet and switched on for everyone. This article explains what happens automatically, what your visitors and callers actually see and hear, and the handful of things the law leaves with you.
The formal, dated version of this — the document to keep in your compliance file or hand to an auditor — is Yaplet's AI Transparency Statement. This article is the practical walk-through.
Who has which duty
The AI Act talks about two roles. Yaplet is the provider: we build the AI features and have to design them so the disclosure is there. You are the deployer: the business using the AI with real customers. Most duties sit with the provider and Yaplet handles them for you; a few sit with the deployer and are listed at the end of this article. Company size does not matter — the same rules apply to a one-person shop and to an enterprise.
What Vex shows in the chat widget
- The conversation header names the sender as an AI assistant ("Vex · AI assistant · replies instantly").
- Before the first message, the empty conversation says in plain words that the visitor is chatting with an AI assistant.
- Every AI reply carries a small AI marker next to the sender name; replies from your human agents don't.
- When one of your agents takes over, the header switches to that person's name and "Live support team member", so the change from AI to human is visible. Read more in Handoff to a human.
The texts follow the widget's language setting (English or Hungarian).
What callers hear on the phone
Every call to an AI phone agent opens with a fixed sentence in the agent's language — in English: "This call is answered by an AI assistant." — and only then your own greeting. In the voice agent editor this appears as a locked AI notice line above your Greeting field. You can't edit, remove or skip it, and callers can't talk over it: even if you allow callers to interrupt your greeting, the notice always plays to the end first. The preview button plays the notice and your greeting together, so what you hear in the dashboard is what callers hear.
The agent is also instructed to answer honestly if a caller asks whether they are talking to a person — it will say it is an AI assistant and carry on helping. See Create a voice agent for the editor itself.
What email and social recipients see
- Email: an answer written by Vex ends with "This answer was written by [your company]'s AI assistant" and, if you allow agent requests, a "Talk to a human" button.
- Facebook Messenger and Instagram: the first AI reply in a conversation starts with a one-line notice that the person is chatting with an AI assistant. It repeats after a pause of more than 24 hours, and again whenever the AI resumes after one of your agents handled the thread — the same moments Meta's own messaging policy asks for.
AI-generated images and text
Images generated in AI Social Posting carry an embedded, machine-readable marker (the IPTC "trained algorithmic media" source type) and are declared as AI-generated to Instagram when Yaplet publishes them; Facebook reads the embedded marker itself. Both platforms show their own "AI info" label. You don't have to do anything for that.
AI-written text — captions, newsletters, articles — is a draft that you review and publish. The AI Act does not require labelling text a human has reviewed and takes editorial responsibility for; see the last section for the one case where it does.
What stays your responsibility
- Leave the notices alone. They can't be switched off, but don't hide them with custom CSS, and don't write a system prompt that tells Vex or a phone agent to claim it is human.
- Don't dress the AI up as a person. Don't upload a real colleague's photo as the bot avatar or write a greeting like "Hi, this is Anna from support" for an AI agent.
- Match the language. The notices follow the widget or voice agent language — set it to the language your customers actually speak.
- Label unreviewed public-interest text. If you publish AI-written news, market or company reports, or public information without a person reviewing it, mark it as AI-generated.
- Label deepfakes. If you publish images, audio or video that show real people, places or events realistically but were made or changed by AI — with any tool — mark them clearly as AI-generated.
- Keep the statement. Save a copy of the AI Transparency Statement with its version date in your compliance records; it is the provider confirmation your procurement or auditor may ask for.
Where to ask
If a customer, auditor or authority asks you for a written confirmation from your AI vendor, point them to the statement above or write to [email protected]. For how Yaplet handles personal data, see DPA and data residency.