It's the question every business deploying an AI phone agent eventually asks: do we actually have to tell people it's not a human? The short answer is that the law is moving firmly toward "yes" — and the smart answer is to do it regardless.
The EU draws a hard line: 2 August 2026
The EU AI Act's Article 50 makes it explicit. From 2 August 2026, providers of AI systems that interact directly with people must design them so that a person is informed they are dealing with an AI — unless it's already obvious. It's a transparency duty, not a high-risk classification, and it carries real teeth: penalties can reach €15 million or 3% of worldwide annual turnover.
If you operate in the EU, or call people who are there, that's not advisory. It's a deadline.
The UK doesn't say it as loudly — but it says it
Post-Brexit, the UK isn't directly bound by the EU AI Act, and there's no single British statute that says "declare your bot." That leads some people to assume the UK is a free-for-all. It isn't.
UK GDPR's fairness and transparency principles — the duty to tell people how their data is being processed under Articles 13 and 14 — make it very hard to argue that secretly running someone through an AI, capturing what they say, is "fair" processing. The ICO's guidance on telephone marketing reinforces the direction: be clear about who is calling and why. The destination is the same as the EU's; the UK just arrives via principle rather than a dated rule.
Two more UK rules worth knowing for outbound AI calls
Disclosure isn't the only obligation once you're dialling out:
- Ofcom's persistent misuse policy requires that any abandoned call plays a short message identifying the caller and offering an opt-out (within two seconds), that a valid caller ID is presented, that abandoned-call rates stay under 3% per campaign in any 24-hour period, and that unanswered calls ring for at least 15 seconds.
- PECR treats automated (pre-recorded) marketing calls differently from live ones — they generally need prior consent. A genuine research or satisfaction survey with no marketing content isn't "direct marketing" at all, but the moment a call sells, the consent rules bite.
The practical answer
You don't need a legal opinion to get this right. Open the call by saying what you are — "Hi, I'm an AI assistant calling on behalf of…". It satisfies the EU's rule, aligns with UK fairness, and — this is the part people underestimate — callers respond better when they're not left wondering. Transparency isn't the compliance cost here. Pretending is.
🎧 We walk through the rules in our audio briefing — listen here.
Related: Compliance & trust · Can an AI run a survey people finish?