Imagine your profile with a label next to the photo: generated by artificial intelligence. Would that stir any feelings?
As of this month, that's no longer a thought experiment. On 2 August, Article 50 of the EU Artificial Intelligence Act came into force: a generated image that looks like a real photo of a real person must be clearly labelled in commercial use. On websites, in employee photos, in advertising.
I'm writing about this because it directly touches my work. And because the technical side of the question is only half the answer.
Why the label matters more than it seems
I've heard the question more than once: is it really worth being photographed when you can generate a photo in five minutes? The same people often add that the generated portrait will look better – and that they're "not photogenic" anyway.
It probably will look better. More regular features, perfect light, skin like crystal.
But better than what? Better than you?
A generated portrait shows not you, but a version you'd like to be. And that's the whole point. The best portrait isn't the one where you look most beautiful. It's the one where people recognise you – and trust you.
What research says about trust
Princeton University research shows a first impression forms in a tenth of a second. In that time the brain decides two things: whether this person can be trusted and whether they look competent. Looking longer barely changes the decision.
Something strange happens with generated portraits. As long as the viewer doesn't know the image is artificial, they rate it even more favourably than a real one. But once they find out, everything flips: surveys show two thirds of HR professionals view a generated candidate portrait negatively.
And one more number that explains why the regulation exists at all: nearly 90% of people want to know whether an image was created by AI.
In other words, an AI portrait works only as long as nobody knows. Since 2 August, in commercial use, it no longer has that luxury.
What is your company under this regulation?
The regulation distinguishes two roles. The provider is whoever builds the AI tool – they carry the obligation to embed a machine-readable mark in the file.
The deployer is whoever uses the tool. If a company or its employees generate images for work purposes, the company is a deployer – and it carries the obligation to visibly label the content.
Personal, non-professional use falls outside the regulation's scope. But if the activity regularly generates economic benefit, the use is considered professional.
What counts as a deepfake?
The term sounds dramatic, but the definition is technical. Content is considered a deepfake when all three conditions are met:
- Resemblance. A high degree of similarity to a depicted person, object, place or event.
- Existence. What is depicted really exists or could exist.
- A false impression of authenticity. The content can mislead the viewer about whether it is real.
In practice this means a generated employee portrait that looks like a real photo triggers the obligation. An abstract illustration, obviously drawn graphics or an image without people does not.
What doesn't need labelling?
The obligation doesn't apply when AI performs an assistive, standard editing function or doesn't substantially change the source data and its meaning:
- Skin retouching, colour and exposure correction.
- Removing dust, defects or a stray hair.
- Noise reduction and sharpening.
The line sits at whether the change creates a false impression of reality. As long as the photo shows what actually happened in front of the lens, there is no obligation.
So a normal photoshoot result, even with full professional retouching, falls outside these rules. Retouching handles light, colour and temporary things. It doesn't change what makes you you.
What should the label look like?
The European Commission is clear here: the label must be understandable and noticeable to a person without any technical tools. Relying only on the machine-readable mark embedded by the AI tool is not enough.
In practice this means a visible caption or symbol next to the image. The EU has published an official icon set for exactly this purpose. The information must reach the person no later than their first contact with the content.
Which brings us back to the question I started with. A label next to a photo is not a punishment. But it is information every visitor to your profile will read before your job title.
Where's the simpler path?
The whole question can be bypassed with one decision. A real photo requires no labelling, triggers no deadlines and leaves no question about what the person seeing it will think.
An hour in the studio – and this whole issue disappears. Nothing to label, nothing to worry about being noticed.
A team photoshoot can be organised in a single day, with the studio coming to your office. The result: photos you can use anywhere, anytime, with no conditions and no explanations.
And most importantly: the person who sees such a photo will recognise you when you meet in person. That is where trust begins.
Frequently asked questions
Do AI photos on a company website need to be labelled?
Yes, if the image is generated or altered by AI and looks like a real photo of a real person, place or event. Abstract illustrations and obviously drawn graphics are exempt.
Does a retouched photo need to be labelled?
No. The obligation doesn't apply when AI performs an assistive, standard editing function or doesn't substantially change the meaning of the source data. Skin retouching, colour correction and cropping fall outside the scope.
Can I use an AI portrait on my LinkedIn profile?
Technically LinkedIn doesn't forbid it, and personal use falls outside the regulation. But surveys show two thirds of HR professionals view a generated portrait negatively, and nearly 90% of people want to know whether an image is artificial. The risk here isn't legal – it's reputational.
Do previously published photos need retroactive labelling?
No. Content created and published before 2 August 2026 doesn't have to be labelled retroactively. The European Commission recommends it, but doesn't require it.
Does the obligation apply to a personal LinkedIn profile?
Personal, non-professional use falls outside the regulation's scope. But if the activity regularly generates economic benefit or is connected to business, the use is considered professional.
Sources: European Commission, “Safer and more transparent AI” (2026-08-02); Commission FAQ on Article 50 of the AI Act; Guidelines on transparency of AI-generated content; Princeton University first-impression research; Ringover recruiter survey; Getty Images transparency research.