Your voice agent has to say it's an AI: what Article 50(1) means for conversational AI vendors from 2 August 2026
If your product talks to people — voice agents for call centers, AI SDRs on the phone, AI receptionists, chatbots — the EU AI Act contains one sentence written specifically about you. Article 50(1): AI systems intended to interact directly with natural persons must be designed so that the persons concerned are informed that they are interacting with an AI system. It applies from 2 August 2026, and the Digital Omnibus did not move that date.
Who owes the disclosure — you, not your model supplier
Article 50(1) assigns the duty to the provider: whoever develops the AI system or places it on the market under their own name. A vendor that ships a voice or chat product under its own brand — even one built on GPT, Claude, or a third-party speech stack — is the provider of that system. "Our LLM supplier handles compliance" is the answer that fails legal review fastest, because the disclosure obligation toward the people your product talks to was never the model supplier's to carry.
The "obviousness" exception, honestly read
The duty does not apply where interacting with an AI is obvious to a natural person who is reasonably well-informed, observant and circumspect, taking into account the circumstances and context of use. Two practical readings for vendors:
- A chat widget labeled "AI assistant" inside a product UI has a strong obviousness argument.
- An outbound phone call with a natural-sounding synthetic voice has almost none. The person answering has no context to infer they're talking to a machine — which is precisely the scenario the paragraph was written for. Voice-first products should assume the disclosure duty applies in full.
What "informed" looks like in practice
Article 50(5) adds the how: the information must be provided in a clear and distinguishable manner at the latest at the time of the first interaction or exposure. For a voice agent that means an up-front disclosure at the start of the call — not a line buried in terms of service, and not a reveal only when the caller asks. Design patterns that survive review: a first-utterance disclosure ("This is an AI assistant calling on behalf of…"), a persistent AI label in chat UIs, and logging that the disclosure fired.
Reach: this is not only an EU-headquarters problem
Under Article 2(1)(a) and 2(1)(c), the Regulation reaches providers placing systems on the EU market regardless of where they're established, and even third-country providers where the system's output is used in the Union. A US voice-AI vendor whose enterprise customer runs EU call flows is in scope. (Full analysis: Does the EU AI Act apply to your non-EU SaaS?)
How the duty shows up in your pipeline (before any regulator calls)
Enterprise buyers now test Article 50(1) directly in vendor assessments: AI-CAIQ and SIG 2026 questionnaires and RFP AI sections ask conversational-AI vendors how end users are informed they are interacting with AI, and what evidence exists. From 2 August 2026 the question stops being hypothetical — and an unconvincing answer parks the deal long before any fine (up to €15M or 3% of worldwide turnover under Art. 99(4)(g); SMEs pay the lower amount under Art. 99(6)).
Answering the questionnaire question well
- Classify — your system interacts directly with natural persons; Article 50(1) applies to you as provider. Say so plainly.
- Describe the disclosure mechanism — where and when users are informed (first utterance, persistent label), with screenshots or call scripts as evidence.
- Address the exception honestly — if you rely on obviousness for some surfaces, say which and why; don't stretch it to phone calls.
- State gaps with a plan — if disclosure isn't shipped on every surface yet, a dated rollout plan beats a hopeful "yes".
- Regulation (EU) 2024/1689, Official Journal — Art. 50(1), Art. 50(5), Art. 2(1), Art. 99, Art. 113 (ELI: data.europa.eu/eli/reg/2024/1689/oj)
- Digital Omnibus: European Parliament position 16 June 2026; Council adoption 29 June 2026 (consilium.europa.eu press releases)
- Cloud Security Alliance — AI-CAIQ; Shared Assessments — SIG 2026 (AI domain)