If your website chatbot answers customer questions, or an AI voice agent picks up your phones, a simple question now has legal weight: does it have to tell people it is not a human? For businesses selling into Europe the answer has been clearer since 2 August 2026, when the EU AI Act's transparency rules started to apply. In the US, a growing set of state laws and a federal ruling on AI voices cover parts of the same ground. The UK has no AI-specific rule, but existing law still matters.
This guide is for owners, operations leads and product teams in the UK, US, Canada and the Gulf who use, or plan to use, chatbots and voice agents with customers. It is a practical overview, not legal advice; rules change and the details depend on your situation, so check with a lawyer before you rely on any of it.
The EU: Article 50 of the AI Act
Article 50 of the EU AI Act sets transparency duties for certain AI systems. The part that matters for most chatbots and voice agents is the first one: providers must design AI systems that interact directly with people so that those people are informed they are interacting with AI, unless that is already obvious to a reasonably well-informed person from the context.
Key points as of November 2026:
- It applies from 2 August 2026. The EU's Digital Omnibus package, approved in mid-2026, delayed the main high-risk obligations (to December 2027 for Annex III systems and August 2028 for Annex I), but law firm commentary such as Jones Walker's July 2026 analysis confirms the Article 50 transparency duties were not delayed.
- There is a narrow grace period, and it is not for chatbots. Providers of generative AI systems placed on the market before 2 August 2026 have until 2 December 2026 to add machine-readable marking of AI-generated content (Article 50(2)). That does not postpone the duty to tell people they are talking to an AI.
- Deepfakes and AI-written public-interest text have their own disclosure duties for deployers (Article 50(4)). A voice agent that imitates a real person's voice falls on the wrong side of this.
- It reaches non-EU businesses. The Act applies to providers placing AI systems on the EU market regardless of where they are established, and to non-EU providers and deployers where the system's output is used in the EU. A UK, US or Pakistani company running a chatbot for EU customers should assume it is in scope.
- Fines are significant. Breaches of Article 50 fall into the tier with fines of up to 15 million euros or 3% of worldwide annual turnover, whichever is higher; for SMEs, the lower of the two applies.
The Commission consulted on draft Article 50 guidelines in May 2026, and a Commission-facilitated code of practice covers content marking. If you sell into the EU, watch for the final guidelines, as they will shape what "obvious from the context" means in practice.
The US: a federal rule for calls and a patchwork of state laws
There is no single US law requiring chatbots to disclose that they are AI. Instead there are several rules that each cover part of the picture.
AI voices on phone calls. In February 2024 the Federal Communications Commission ruled that AI-generated voices count as "artificial or prerecorded voice" under the Telephone Consumer Protection Act. It did not ban the technology, but calls using an AI voice need the same consent as robocalls: prior express consent for non-marketing calls to mobiles, and prior express written consent for marketing calls. The calls must identify the business and give a phone number, and marketing calls need an automated opt-out. This matters most for outbound calling; an inbound AI receptionist avoids most of it.
State chatbot laws. Several states have passed laws in the past two years. According to summaries by Cooley (October 2025) and Wiley (June 2026):
- Utah requires a business using generative AI with consumers to disclose it when a consumer clearly asks whether they are dealing with a human, and to disclose it up front in regulated occupations and "high-risk" interactions, such as those involving sensitive personal data or financial, legal or health services.
- Maine requires notice that the consumer is not dealing with a live human where a reasonable consumer could not tell the difference, enforceable under its unfair trade practices law (in effect since September 2025).
- California and New Jersey prohibit using bots to deceive consumers in online commercial transactions without disclosure.
- California's SB 243 and New York's AI companion law set stricter rules for "companion" chatbots, including repeated reminders that the user is talking to AI. Both have exceptions or carve-outs that are likely to cover bots used only for customer service, but check how your bot is classified.
More states have companion and mental-health chatbot laws coming into force in 2026 and 2027. If your bot handles health, money or anything emotionally sensitive, take specific advice.
The UK: no AI disclosure law, but existing rules apply
The UK has no statute that requires a chatbot to announce itself. That does not mean anything goes:
- Consumer protection law prohibits misleading consumers. Letting customers believe they are speaking to a person, when that affects their decisions, carries risk.
- UK GDPR requires you to tell people how their personal data is used, including chat logs, call recordings and transcripts, and gives people rights around solely automated decisions with significant effects.
- Automated marketing calls need prior consent under the Privacy and Electronic Communications Regulations (PECR).
In practice, if you serve customers in both the UK and the EU, the simplest approach is to meet the EU standard everywhere.
Canada and the Gulf
Canada's proposed federal AI law (AIDA) did not pass, so there is no national chatbot disclosure rule; privacy law (PIPEDA, and Quebec's Law 25 for automated decisions) still applies to the data your bot collects. In Saudi Arabia and the UAE, the main obligations come from data protection laws (Saudi PDPL, UAE PDPL and the DIFC and ADGM regimes) rather than chatbot-specific rules. A clear disclosure is still good practice and costs nothing.
What good disclosure looks like
Disclosure does not have to make a bot awkward. A short, friendly line at the start is usually all it takes.
- Say it at the start. "Hi, I'm the virtual assistant for Smith & Co" in a chat window, or "You're speaking to our AI assistant" in a voice greeting.
- Keep it visible. A label on the chat widget, not just a line in the privacy policy.
- Answer honestly when asked. If a caller asks "Am I talking to a real person?", the agent must say no. Test this in your scripts.
- Offer a way to a human. A button, a callback or a transfer, especially for complaints and anything sensitive.
- Do not clone real voices of staff or public figures without clear consent and a legal check.
- Tell people about recording and data. Who processes the audio and transcripts, where, and for how long.
- Log the disclosure. Keep a record that the greeting was played or shown, in case you need to show compliance later.
Building it in from the start
Disclosure is cheapest when it is part of the design. When we build a chatbot or voice agent, the greeting, the "are you human?" answer, the human handover route and the data notice are written into the agent's instructions and tested alongside everything else. For Brixven, an agency serving small businesses in the UK and Ireland, we built a Vapi voice agent that answers inbound calls and books appointments, connected to n8n workflows for email follow-up and CRM updates.
If you are choosing between an off-the-shelf AI receptionist and a custom agent, our guide to AI receptionists for small businesses compares the two. You can also see what we build on our AI solutions page.
If you want a chatbot or voice agent that is useful to customers and set up properly for the markets you sell into, talk to us.
Need help with this? See our AI agent and voice AI development page, or book a free discovery call and a senior engineer will reply within 24 hours.