The EU AI Act probably does not apply to you. Do it anyway.
On 2 August, a chunk of the EU AI Act became enforceable. Specifically Article 50, which covers transparency, and which is the part most likely to touch an ordinary business.
You will have seen people talking about it, usually with a large number attached. So let me set out what actually changed, whether it applies to you, and why I think you should follow it regardless of the answer.
Usual caveat: I am not a lawyer and this is not legal advice. If you think you are genuinely in scope, go and get proper advice rather than taking mine.
What Article 50 actually requires
Three things, in plain terms.
If somebody is talking to an AI, tell them. Any system that interacts with people has to make clear it is an AI, unless it is obvious from the context. That covers chatbots, and it covers AI phone systems.
If you are analysing emotions, say so separately. Emotion recognition and biometric categorisation carry their own disclosure duty. If your system is assessing how somebody sounds or feels, that is a distinct thing to declare.
If you publish AI-generated images, video or audio of people, label it. Deepfakes and synthetic media depicting real people need to be marked. There is also a machine-readable marking requirement for generative systems, with a slightly later deadline of 2 December for things already on the market.
The penalties sit at up to €15 million or 3% of worldwide annual turnover, whichever is higher.
And a detail worth knowing, because it caught people out: this was not delayed. The Digital Omnibus pushed the high-risk AI rules back to December 2027, and a lot of businesses read the headline and stood down. Article 50 was not part of that deferral. It landed on schedule.
Now the question nobody is asking
Here is what has been missing from most of the coverage I have seen.
Does this apply to a business in Lancashire?
The UK is a third country now. There is no automatic obligation. But the Act reaches beyond the EU’s borders, and the test is not where your business is. It is where the output of your AI system is used.
If your AI system’s output is used by, or affects, somebody in an EU member state, you are in scope regardless of where you are registered. It works much like GDPR, except the threshold is arguably lower: GDPR asks whether you are targeting EU residents, whereas this asks whether the output is used there.
So, practically:
A plumber in Poulton with an AI phone system, taking calls from Lancashire. Almost certainly out of scope. Your output is used in Poulton.
A Blackpool hotel with an AI booking assistant. Rather less clear, and this is the one worth thinking about. If somebody in Dublin or Berlin enquires about a room and gets an answer from your AI, the output has been used in the EU. Ireland is under an hour away by plane and a meaningful chunk of Fylde coast tourism.
Anybody selling online across Europe. In scope, and you probably already knew that.
I am not going to pretend the enforcement risk against a small Lancashire hotel is high. Realistically, regulators have larger things to look at. But “unlikely to be prosecuted” is a different statement from “does not apply,” and it is worth knowing which one you are relying on.
And the UK is not a free-for-all either
Worth saying, because “the EU Act does not apply to me” gets heard as “there are no rules.”
The UK has no equivalent single AI law. It has taken a sector-led approach, with existing regulators applying existing rules within their own patch. But those existing rules have not gone anywhere.
Misleading people about who or what they are dealing with is already a problem under consumer protection and advertising rules. The ICO already has firm views about automated decision-making and transparency. None of that needed a new AI Act.
So the honest position is not “the EU rules do not apply, carry on.” It is “the specific EU rules probably do not apply, and the general obligation to not mislead people never went away.”
Why I would follow it regardless
Here is my actual argument, and it has almost nothing to do with the law.
People feel had when they find out.
I have watched somebody realise, halfway through a conversation, that the friendly voice they have been chatting to for two minutes was not a person. The reaction is never neutral. It is a small, sharp sense of having been made a fool of, and it attaches to the business rather than the technology.
Whereas being told up front costs you almost nothing. Most people are entirely happy to talk to an AI if it can actually help them, and if they know what it is. What they mind is the discovery.
The voices are good enough now that this genuinely matters. A few years ago disclosure was almost redundant because nobody was fooled. That is no longer true, which is precisely why the rule exists.
And there is a straightforwardly commercial point. If your whole pitch is honesty, you cannot have an exception for the robot. I write a great deal about being straight with people. It would be a strange business that argued for transparency in pricing and then quietly let its phone system pretend to be a bloke called Dave.
What that looks like in practice
It is genuinely a sentence.
For an AI answering enquiries: say so at the start. “Hello, you are through to an automated assistant for [business], I can take your details and get somebody to ring you back.”
That is it. No lengthy disclaimer, no legalese. Just not pretending.
For AI-generated images of people in your marketing: label them. If it is a photograph of a person who does not exist, say so. This does not apply to a logo or an illustration, and nobody sensible thinks it should.
And the general rule I would apply, which covers most situations without needing to check anything: if somebody would be annoyed to discover it later, tell them now.
What we do
Since I build AI response systems into client sites, I should say what I actually do rather than just recommending things.
Every one of them identifies itself as an AI assistant at the start of the conversation. Not because a regulator told me to, but because the first time somebody feels tricked by it, the business it belongs to has lost more than it gained.
None of my clients are currently exposed to the EU rules as far as I can tell, and I have checked the ones where it is a live question. But the disclosure would be there either way, because it is the right way round.
The short version
The EU rules landed on 2 August and were not delayed. They probably do not apply to you if your customers are all in Lancashire, and they might if some of them are not.
But the underlying instruction, tell people when they are talking to a machine, is not really a legal question. It is the same thing you would do if it had never been written down.
Say it is an AI. Everybody relaxes, including you.
Want to talk about doing any of this properly? Get in touch, or email hello@inkon.co.uk. I read every message myself, and it is definitely me.