What the UK rules actually say about AI voice agents
Everybody selling these is confident about the legal position. The published guidance is narrower than most of them imply, and on the question that matters most it does not clearly say either way.

Umer Ejaz Chaudhry, Cadral AI ·
If you are considering an AI agent that makes calls, you will have been told it is fine. You may not have been told what the actual rules are, or which parts of them nobody can currently answer.
What follows is what the Information Commissioner’s Office publishes, read from its own guidance on 26 August 2026 rather than remembered. Where it runs out, we say so instead of filling the gap.
1. The rules for recorded calls are stricter than for live ones
This is the part most people get wrong, and they get it wrong in the expensive direction. Automated marketing calls sit under their own regulation and the bar is higher than for a person dialling.
The requirement is specific consent: consent that covers automated calls particularly. General consent to marketing does not carry over, and neither does consent to receive live calls.
Somebody agreeing to be phoned has not thereby agreed to be phoned by a machine (ico.org.uk, Guide to PECR, telephone marketing, checked 26 August 2026).
There are presentation obligations too. An automated call has to identify who is calling and give a contact address or freephone number, and the calling number has to be displayable.
2. Live calls have their own, different rules
Live marketing calls sit under separate regulations, and the mechanism there is mostly screening rather than consent. You must not call somebody who has told you they do not want your calls.
You must not call a number registered with the Telephone Preference Service, or its corporate counterpart the CTPS, unless that person has specifically consented.
Which means call lists need screening against both registers, and the corporate one matters as much as the individual one for anyone selling business to business (same source, same date).
3. The guidance is currently under review
The ICO’s own page carries a notice that this guidance is under review following the Data (Use and Access) Act and may change.
That is not a technicality worth skipping over. Anything load-bearing should be checked against the text as it stands when you make the decision, rather than against a blog post, including this one.
4. The question the guidance does not answer
Here is the part that matters, and the part we have not seen anyone in this market put in writing. The automated calling rule is written around what the guidance describes as an “automated dialling system that plays a recorded message”.
That phrasing predates the current generation of voice models by a long way, and it describes a specific thing: a machine that dials and plays a recording.
A generative agent holding an actual conversation, listening, answering an unexpected question and changing course, is not obviously that. It is also not obviously a live call, because there is no live human on the line.
So which set of rules applies is a real open question. It may be treated as caught, or it may not, and that may be settled by the review underway or by somebody being made an example of. The guidance does not yet say.
5. What this means in practice
Any supplier telling you confidently that their AI caller falls outside the automated calling rules is guessing, and guessing in the direction that suits them. That is a poor basis for a decision whose liability sits with you, not with them.
The way through is to stay on the side of the line where the ambiguity never needs resolving. Calls to people who have asked to hear from you, about the thing they asked about, are a different proposition from calls to strangers selling something.
Confirming an appointment somebody booked, ringing back a person who left a message, chasing a document that is holding up a job they commissioned: those are not the calls the marketing rules are aimed at.
Cold outbound is where the question bites, so we build every outbound campaign to the stricter automated-calling standard, whichever way the guidance is read: consent that specifically covers automated calls, and lists screened against the TPS and CTPS before and during every campaign.
What this does not prove
This is not legal advice, and we are not solicitors. It is a description of published guidance, read on a named date, plus a plain statement of where that guidance runs out.
If your intended use is anywhere near the line, that is a conversation for somebody qualified to give you an answer you can rely on.
It is also a post about the rules, not a record of any calling programme. Knowing what the guidance says and running a campaign inside it are different things, and this post claims only the first.
And the guidance is under review, so this post has a shelf life. It is dated at the top for that reason. Check the source rather than trusting the summary.
Read something here you want running in your business? Send us the brief and we will come back with a fixed price, or book twenty minutes and talk it through.
Tell us the job. We will tell you the price.
Twenty minutes and a fixed quote in writing. You bring the brief and the brand. Everything after that is us.
Not ready to talk? Run your own numbers on fifteen free tools, no email asked.