AI Act Article 50: Does an AI Phone Agent Have to Say It Is AI?

Since 2 August 2026, Article 50 of the EU AI Act, Regulation (EU) 2024/1689, has applied across the Union. For any business that runs an AI phone agent, or is about to, the question is simple: does the agent have to tell callers it is AI? The short answer is yes, and it has to do so at the start of the call. Below we walk through what the regulation actually asks for, who carries the obligation, and what to check before you put an AI phone agent in front of your customers.

What Article 50 of the AI Act says about AI phone agents

Article 50 is the transparency part of the regulation. Its first paragraph says that AI systems intended to interact directly with natural persons must be designed so that those people are informed they are interacting with an AI system. An AI phone agent that answers your customers' calls is exactly that kind of system.

Paragraph 5 of the same article covers the how. The information has to be clear and distinguishable, and it has to be given at the latest at the time of the first interaction. On a phone call, that puts the disclosure at the opening of the conversation, not halfway through and not in a terms-of-use page on your website.

There is one exception: no disclosure is needed when it is obvious to a reasonably well-informed person that they are talking to AI. On a phone call with a natural, human-sounding voice, that exception is hard to rely on. The more natural the agent sounds, the less obvious it is that a machine is on the line, and the more an explicit disclosure matters.

A common question is whether the Digital Omnibus simplification package changed any of this. According to the analyses published so far, the Omnibus mainly pushed back the obligations for high-risk systems, while the duty to tell people they are talking to AI has applied as planned since 2 August 2026.

How it works on a real call

Transparency is not just one sentence at the start of a call. It is a way of designing the conversation so the caller always knows who they are dealing with and always has a route to a person. In Voxia, that looks like this:

• An audio notice on every call, stating that the conversation is handled by a digital assistant with automated voice processing, played before any audio is sent to a speech-recognition provider.
• The agent never presents itself as human, not even when a caller asks directly, or when the conversation flows so naturally that they forget.
• Handover to a person when it is needed, as an attended transfer: the agent first briefs your staff member with a spoken summary, then connects the caller.
• Call recording with a legal notice, wherever you switch recording on, so the caller is also told about that part of the processing.
• A full audit trail and call history in the dashboard, so you can show what happened on any given call if you are ever asked.

IT & ComplianceLearn more →

Who is responsible: the provider or the business?

The AI Act distinguishes the "provider", who develops the system, from the "deployer", the business that uses it. The Article 50(1) duty sits primarily with the provider: the system has to be built to disclose. That does not mean a business can ignore the issue. If the system you use lets the disclosure be hidden or switched off, the consequences land with you and with your customers.

Two more points are worth keeping in mind. First, the AI Act does not replace the GDPR: recording, processing and retaining personal data are still governed by it. Second, most uses of an AI phone agent, such as appointments, information or order status, fall under the transparency obligations rather than the stricter high-risk regime. Some sectors, however, such as creditworthiness assessment or risk pricing in life and health insurance, are named in the regulation as high-risk and need a separate assessment.

The regulation sets fines for breaches of the transparency obligations of up to 15 million euros or 3% of worldwide annual turnover, with a lighter cap for SMEs. The bigger cost is usually a different one: a customer who finds out afterwards that they were talking to a machine, without anyone telling them, rarely trusts you again.

What to check before you switch on an AI phone agent

Before an AI phone agent starts answering your customers, ask the provider whether the AI disclosure happens on every call and right at the start, and whether it can be turned off. Ask what happens when a caller asks for a person, and whether the handover comes with a summary or leaves the customer repeating everything from scratch.

Check where the data is processed, too. Voxia can be deployed in the cloud or on-premise; on-premise, the data never leaves your organisation. In sectors with heavier requirements, such as insurance, financial services or healthcare, that option makes the conversation with your compliance officer much simpler.

Finally, make sure the disclosure lines up with your other channels. The privacy policy on your website, the recording notice and the message on the call should all say the same thing. That consistency is what makes transparency credible, both to your customers and to anyone who audits you.