AI Transparency Disclosure

SwiftGuard provides an AI voice agent that answers telephone calls for plumbing and heating businesses. This page says plainly what that means: when you call a business that uses SwiftGuard, you are talking to artificial intelligence, and the agent tells you so at the start of the call. It also sets out how you are told that - which is what Article 50(1) of the EU Artificial Intelligence Act (Regulation (EU) 2024/1689, the "AI Act") requires of the provider of a system like ours, and those rules have applied since 2 August 2026.

This document covers that one thing: how a caller is told they are speaking with AI. It is not a complete account of the AI Act or of every duty that Regulation places on us, and it does not claim to be. What happens to the information from a call is in our Privacy Policy.

In this document, you means anyone reading it - a caller, or a business that uses SwiftGuard. Where a point concerns only one of the two, it says so.

In short

• Every call the agent answers opens with the agent saying it is an AI assistant for the business you called, and that the call is recorded - before you are asked for anything. That opening line is fixed wording, not something the AI composes. If you speak over it or the line is noisy, or a technical fault drops the connection at the start of a call, you may hear part of it or none of it - and you can ask at any point instead.
• The agent is instructed never to pretend to be a person, and to say it is an AI assistant whenever you ask whether you are talking to a human or a machine.
• If the conversation switches to another language mid-call, the agent is instructed to say the same thing again in that language. That is an instruction the agent follows, not fixed wording like the opening line, so it is not guaranteed in the way the opening line is.
• The disclosure is built in and cannot be switched off - not by us, and not by the business using SwiftGuard.
• You never have to keep talking to it: you can ask it to take a message for the business's staff, or hang up and contact the business another way.
• Questions: [email protected].

1. The AI system

SwiftGuard's voice agent is a conversational AI system. It answers the phone for the business you called, speaks with you in natural language with a synthetic voice, and takes down your details and what you need. On a call it can book a job, move or cancel a job you already have, take a message for the business's staff, and write down or correct the details the business holds for you. Before it cancels a booking, signs you up, or changes the details the business holds for you, it checks that the phone is yours: it texts a short code to the number you are calling from and asks you to read it back. That check happens once on a call: the same code covers anything else of that kind you ask for on the same call, so you are not asked to read a fresh one back for each of them. On a long call that check can time out, and you may be asked to read a new code back. Booking a job is not checked that way: a caller the business already knows is recognised from the postcode and house number the business holds for them. Calls are recorded and transcribed so the business can see what was agreed; what data that involves, how long it is kept and the rights you have over it are in our Privacy Policy.

2. How callers are told

The agent's very first line, on every call, says two things in the language of the greeting: that you have reached the AI assistant for the business, and that the call is recorded and written down. That line is fixed wording, the same on every call the business takes. Two things can come between you and it. If the agent hears you start speaking - or hears enough noise on the line to take it for speech - it stops, and the rest of the line is not played to you. And if a technical fault drops the connection at the start of the call, the call is reconnected and carries on, but the opening line is not spoken again - so depending on when the fault happened you may have heard part of it, or none of it. If either happens, ask - the rest of this section is how. Beyond that opening line:

• If the agent switches to another language during the call, it is instructed to say again - in the new language - that you are speaking with an AI assistant and that the call is recorded, so a caller who did not follow the opening greeting is told in a language they do follow. That repeat is an instruction the agent is given rather than fixed wording, so unlike the opening line it is not guaranteed.
• Asked at any point whether it is a person, the agent is instructed to say it is an AI assistant, and never to pretend to be human.
• If you object to being recorded, the agent answers honestly: it cannot switch the recording off, and if you would rather not be recorded you can hang up and contact the business another way. It is instructed not to argue with you or to try to talk you out of it.
• If you would rather deal with a person, ask the agent to take a message - it is passed to the business's own staff, so a person can get back to you.

The announcement comes before anything is asked of you. If you happen to speak over it, or a fault costs you it, you can ask the agent at any point in the call whether you are talking to a person, and it is instructed to tell you.

A telephone call carries only sound - there is no screen to show you a notice on - so everything above is spoken to you out loud.

3. Who is responsible for what

The AI Act splits duties between the provider of an AI system (who builds and supplies it) and the deployer (who uses it).

• SwiftGuard is the provider. SwiftGuard is provided by SwiftGuard AI, our registered name with the Netherlands Chamber of Commerce (KvK) under number 75538288 - a sole proprietorship (eenmanszaak) established in the Netherlands. We develop the voice agent and supply it as a service, so Article 50(1) - designing the system so that people are informed they are interacting with AI - is our duty. We meet it with the built-in spoken disclosure described above: it is part of the system for every business, and it opens the call in the language the call is answered in, subject to what section 2 says about what can interrupt it.
• The business whose calls the agent answers is the deployer. Because the disclosure is built into every call and cannot be disabled, the business does not need to configure anything for its callers to be told they are talking to AI. The business keeps its own duties under data protection law as the controller of its callers' data - our Data Processing Agreement covers that split, and section 8 of our Terms of Service covers call-recording consent where a country's law requires more than an announcement.

4. What the system does not do

• It does not analyse your voice. It does not work out how you are feeling from the sound of it, and it does not identify or categorise anyone from their voice or from any other biometric data. It works from your words: what you say is turned into text, and the AI reads that text - so if you tell it something is urgent, or that you are worried about it, it will respond to that and may mark the job urgent for the business, in the same way a person reading your message would. It does check who it is speaking to - from the details you give it, and from a code texted to the number you are calling from - but never from the sound of your voice.
• It does not take decisions with legal or similarly significant effects on callers. It carries out what you ask of it on the call: booking, moving or cancelling a job, taking a message, and writing down or correcting your details. It does not make the business's decisions for it - what a job will cost, who is sent out, or whether the work is taken on. It does take a few decisions of its own. For example: if a caller is abusive it gives one warning, then offers to take a message and ends the call; if you report a gas smell, a suspected leak or a carbon monoxide alarm, it stops booking and tells you to leave the building and call the emergency services; and if it cannot see the number you are calling from, it will not book or change anything, and takes a message instead so the business can ring you back.
• It does not copy anyone's voice, including yours. The voice you hear is synthetic: one of the standard voices our speech provider offers, chosen for a clear and professional tone rather than to resemble any individual, and we do not build voices from recordings of real people. The transcript and summary of a call that the business sees are generated automatically, as the record of a conversation the caller knew was with an AI.

5. Changes and contact

When the product or the law changes materially, we publish a new version of this document under our usual revision process; the date at the top tells you which version you are reading. Questions about this disclosure, or about how the agent presents itself on a call: [email protected].