AI Act ready out of the box
The EU AI Act's transparency duty for AI chat and voice assistants (Article 50) has applied since 2 August 2026. Here's what it actually asks for, and exactly what CustomerEagle does about it on each channel — verified against the running code, not a marketing claim.
August 21, 2026
The short version
- Article 50's transparency obligation has applied since 2 August 2026.
- A support chatbot like CustomerEagle's sits in the AI Act's "limited-risk" category: it must disclose that a customer is talking to AI, not clear a high-risk conformity process.
- The disclosure has to reach the person at, or before, their first interaction with the system — a line buried in a footer or a terms page doesn't count.
- Non-compliance is enforced under the Act's general penalty regime: fines of up to €15 million or 3% of worldwide annual turnover, whichever is higher.
01What Article 50 actually asks for
Article 50 of the EU AI Act is narrow and specific: if a person is interacting with an AI system rather than a human, they have to be told — clearly, and before or during the first exchange. It does not require the AI to be perfect, explainable in detail, or independently audited; the obligation is disclosure, not certification. For a support chatbot or an AI-answered phone call, that means the customer needs to know, in plain language, that they're talking to a machine.
- Applies to any AI system that interacts directly with a natural person — chat, email and voice all count.
- The disclosure must be clear and understandable to the person receiving it, in a format a layperson notices, not a technical footnote.
- It does not apply where the AI use is "obvious" to a reasonably well-informed person — a defence CustomerEagle deliberately doesn't rely on, since every channel below discloses explicitly instead.
02Who this applies to, and what's at stake
It applies to you if you deploy an AI chatbot, an AI-drafted auto-reply, or an AI-driven phone call to talk to your own customers in the EU — regardless of your company's size or where you're headquartered. Enforcement runs through each EU member state's market surveillance authority, and the penalty ceiling (Article 99) is shared with the Act's other obligations: up to €15 million or 3% of global annual turnover, whichever is higher, for the worst infringements. Getting the disclosure right is a small amount of engineering for a very large amount of downside avoided.
What CustomerEagle does, channel by channel
Not a promise — a description of the running code, audited line by line.
Web chat widget
Always onEvery AI-authored message in the chat widget carries a visible "AI" chip next to the assistant's name — including streamed answers, low-confidence handoff cards, and every reply generated by a custom flow. There is no setting to remove it: it renders regardless of what you name your bot.
Voice (AI phone calls)
MandatoryEvery AI-driven call — whether it runs a code-template script or a tenant-authored scenario — opens with a spoken disclosure naming your company, rendered from your verified business identity before the conversation starts. Free-form AI calls with no opening don't exist in the product, and a tenant must accept the AI-disclosure terms before any AI call can be placed at all.
WhatsApp, Messenger, Instagram & Telegram
On by defaultBy default, AI replies on these channels are sent under a "{workspace} AI" sender name. You can rename your assistant in Settings — if you do, keep "AI" (or your local-language equivalent) in the name so the disclosure stays intact under Article 50.
Email auto-replies
Configure itAn AI-drafted or AI-sent email reply today carries your workspace's normal branding and reply-to address — it doesn't yet stamp an automatic AI label into the message body. If you turn on AI email replies, add a one-line disclosure to your reply signature (Settings → Email) so an emailed answer is as clearly labelled as your chat and voice channels already are.
The rest of the compliance picture
EU hosting
Every CustomerEagle workspace runs on EU-based infrastructure today. Data is encrypted in transit and at rest, and each workspace's data is stored separately from every other customer's.
Data Processing Agreement
A GDPR Article 28 Data Processing Agreement is available to every customer — read it, or download it for your own records, with no sales call required.
Read the DPAHuman sign-off on AI actions
The AI never executes a refund, an order change or another consequential action on its own. Every AI-proposed action that would change something for a customer is queued for a human teammate to approve — or, only within limits you configure yourself, auto-approved under a bounded policy your team controls. The model itself never gets the keys.
The human-handoff guarantee
Every channel is built to step back for a person: web chat surfaces a "talk to a human" option the moment its confidence drops, ambiguous or sensitive email categories are routed to your team instead of auto-answered, and every AI phone call can transfer to a live agent. The AI is never the only door.
Article 50, in plain answers
The questions we get asked most about the EU AI Act and AI customer support.
Does the EU AI Act apply to a customer-support chatbot?
Yes, but narrowly. A support chatbot falls under Article 50's transparency obligation — you must disclose that the customer is talking to AI — not the Act's high-risk category, which covers things like biometric identification or credit scoring. There's no conformity assessment or CE marking to do for a support assistant; there is a disclosure duty.
When did this become enforceable?
Article 50's transparency obligations have applied since 2 August 2026. If your AI assistant went live before that date, the obligation still applies to it now.
What counts as a valid AI disclosure?
It has to be clear, in plain language, and reach the person at or before their first interaction — not buried in a terms page or a footer. "You're chatting with an AI assistant" next to every AI message, or a spoken opening line on a call, both qualify; a one-time popup a customer can dismiss and never see again is the risky end of that spectrum.
What happens if we don't disclose?
Enforcement sits with each EU member state's market surveillance authority, and the penalty ceiling under Article 99 is shared across the Act's obligations: fines of up to €15 million or 3% of your organisation's worldwide annual turnover, whichever is higher, for the most serious cases.
Does CustomerEagle handle this for us automatically?
On web chat and voice, yes — the disclosure is built into the product and cannot be switched off. On WhatsApp, Messenger, Instagram and Telegram, the disclosure ships on by default through the assistant's sender name, and stays intact as long as you keep "AI" in that name if you customise it. On email, you'll want to add a short disclosure line to your own reply signature — see the channel table above.
Is this legal advice?
No. See the note at the bottom of this page — talk to your own counsel to confirm what Article 50 and the wider AI Act require for your specific business.
Related reading
This page describes the product. For the legal terms governing how CustomerEagle processes your customers' data, read the Data Processing Agreement, or see the full security and hosting picture.
See it disclose itself
Every AI reply your customers see already carries the "AI" label described above — try the widget on your own site before you take our word for it.
Questions about your specific setup? Talk to us
This page is a plain-language product overview, not legal advice, and it isn't a substitute for your own compliance review. The EU AI Act's requirements depend on your specific use of AI and your organisation's circumstances — confirm your obligations with your own legal counsel.