AI Act

The European AI Regulation, commonly known as the AI Act, sets rules for offering and using AI systems in the European Union. The rules become stricter as the risk increases, and separate transparency rules apply to systems that generate content resembling real people.

Clonefy creates videos in which a digital version of a real employee speaks. That falls squarely within those transparency rules. This page explains what that means, what we have built into the product for it and which part lies with you as a customer.

The AI Act divides obligations across different roles. For our services, that division looks like this:

Clonefy is the provider

we place the AI system on the market under our own name. That remains the case when underlying models run at specialised suppliers: towards you, we are responsible for the system.

You are the deployer

you use the videos within your own organisation, on your own website, intranet or learning environment. Part of the transparency obligation is tied to that use and therefore lies with you.

The clone person is the data subject

the employee whose image and voice are used. Their consent is the foundation of everything we make.

We have set up our product so that you can meet your part without building anything yourself. How that works is explained in sections 4 and 5.

Clonefy is not a high-risk system within the meaning of the AI Act. We do not decide on access to education, employment, credit, insurance, social benefits or the administration of justice, and we are not used as a safety component in a product. What we do is translate explanations and bring them to the screen.

We also do not engage in any of the practices prohibited by the AI Act. Specifically, this means:

• we do not perform biometric identification: our systems do not recognise individuals and do not search for anyone in footage;
• we do not perform emotion recognition in the workplace or in education;
• we do not engage in social scoring or the assessment of people;
• we do not build facial recognition databases and do not scrape images from the internet;
• we do not use techniques intended to influence behaviour subconsciously.

What does apply to us are the transparency obligations for systems that generate synthetic images and audio. These have applied since 2 August 2026, and they are what the rest of this page is about.

Under the AI Act, a video in which a digital version of an existing employee speaks is a deepfake. The word sounds heavier than it is here: it simply means that the content resembles a real person and must therefore be recognisable as AI-generated. We have arranged this in two ways.

4.1Visible to the viewer

Our player shows that the video was made with AI. This label is there by default and does not need to be switched on or off per video. The viewer does not have to look for it and does not need any technical tool to see it.

4.2Readable by machines

The generated files also carry a marking that software can recognise, so that platforms and other systems can establish that the content has been artificially generated. That marking travels with the file.

4.3Conversations with a clone

At the moment, we do not offer a clone that holds live conversations with viewers. As soon as we do, the viewer will be shown at the start of that conversation that they are talking to an AI system and not to a human. We will update this page at that time.

As a deployer, you have your own obligation to disclose that the content was made with AI. In the vast majority of cases this is already taken care of because you use our player: the label is built in and travels with every embed.

There are two situations in which you need to take action yourself:

• if you download a video and publish it outside our player, for example in a presentation or on your own channel, make sure yourself that there is a visible statement that the video was made with AI;
• if you use our player on your own website, intranet or learning environment, the environment around it remains yours. Also think of consent for non-essential cookies and similar techniques on those pages.

You also remain responsible for the content of your scripts. What the clone says is up to you. We do not generate claims, instructions or advice of our own accord.

We only create a clone of someone who has given prior consent for this. That consent is specific and in writing, and covers making the recordings, building the clone and using it for the agreed purposes.

Consent can be withdrawn. If that happens, we no longer make new videos with that clone. What happens to material that has already been made is discussed case by case, because agreements with the employer and ongoing publications are also involved.

If we doubt whether consent exists, is valid or is broad enough, we put the work on hold until this is clear. This is not a non-binding policy: it is also set out in our general terms and conditions.

There is always a human between the system and publication. You supply the script, you see the result and you decide whether it goes out. Nothing is published automatically.

This is not a formality, because AI output can contain errors. Translations can shift a nuance, the pronunciation of technical terms or proper names can differ, and timing or intonation can turn out differently than intended. Therefore always have translations about safety, health or compliance checked by someone who has a good command of the language.

We do not take decisions about people with this system. No one is assessed, ranked or selected.

A customer’s recordings, voices, scripts and videos remain that customer’s and are not used for other customers. We do not use this material for general model training or as an external showcase, unless separate and demonstrable consent has been given for this.

To deliver our services, we work with specialised suppliers of video, voice and translation technology and with parties for hosting and storage. Agreements on confidentiality and processing have been made with them. Which categories these are is set out in our privacy policy; a current list of our subprocessors is available on request.

The AI Act expects organisations that offer or use AI to ensure that the people working with it understand what the system does and where its limits lie. This obligation therefore also applies to you as a customer.

We address this during onboarding. It covers how a clone is created, where the output is reliable and where it is not, why a script must be checked before it goes out, and what you should tell your employees about the use of their image and voice.

The AI Act enters into force in stages. For our services, these are the relevant dates:

We comply with the obligations that are currently in force and are preparing for those that apply later. If the product or the legislation changes, we update this page and increase the version number.

This page describes how we handle AI. How we handle personal data is set out in our privacy policy, and the arrangements on delivery, use and liability are set out in our general terms and conditions. If a contractual arrangement with a business customer deviates from this page, the contractual arrangement applies to those services.