Unlabeled AI photos: who faces a fine of up to €15 million under the AI Act
EU Requirements
Do you use artificial intelligence to create product images, texts, videos, or to communicate with customers? From 2 August 2026 , new transparency requirements for AI content will start applying to many companies operating on the EU market.
It is not enough to tell users that artificial intelligence is being used. In a number of cases, the content must have machine-readable labeling, and responsibility may be shared between the AI system provider, the agency, the online store, and the company that publishes the material.
We explain who is covered by Article 50 of the EU AI Act, which materials need to be labeled, and why a simple “created with AI” caption may not be enough.
What happens on 2 August 2026
From that date, the transparency requirements set out in the EU Artificial Intelligence Regulation will start to apply. In particular, Article 50 governs informing users and marking certain materials created or modified with AI.
The requirements may affect companies that:
- use chatbots and AI assistants to communicate with customers;
- generate or substantially modify product images;
- create audio, video, and text with AI;
- publish deepfakes or realistically altered materials;
- use AI-generated texts to inform the public on matters of public interest;
- provide AI systems to other companies or users.
The requirements may also apply to companies outside the European Union if their AI systems, or the outputs of those systems, are used in the EU.
Who is covered by Article 50 of the AI Act
Obligations depend on the company’s role. The AI Act distinguishes between providers of AI systems and organizations that use such systems in their work.
AI system providers
Providers of systems that generate synthetic text, images, audio, or video must make it possible to determine that the output was created or modified by artificial intelligence.
Such labeling must be machine-readable and, as far as technically possible, effective, interoperable, robust, and reliable.
Companies using AI content
For businesses, specific requirements arise, for example, when publishing deepfakes and certain AI-generated texts intended to inform the public.
In addition, a person must understand that they are interacting with artificial intelligence, unless this is obvious from the context. This is especially important for chatbots, virtual advisors, and automated customer support.
Not sure whether Article 50 applies to your business?
We will review where and how your company uses AI, identify potential transparency requirements, and prepare a clear list of necessary changes.
Do you need to label every image and every text
Not every use of artificial intelligence automatically means that a visible “created by AI” label must appear next to the material.
What matters is how the material was created, the degree of human involvement, the nature of the changes, the purpose of the content, and your company’s role in publishing it.
Example: an online store
Imagine a store using AI to prepare a product card. Several situations are possible:
- a photographer removed the background from a real product photo using an AI tool;
- AI adjusted the lighting or performed standard retouching;
- the product was placed in a fully generated interior;
- the product image was created without a real photograph;
- AI substantially changed the appearance, color, size, or characteristics of the product.
These scenarios cannot automatically be treated as the same. Standard editing of an ancillary nature is different from creating or substantially modifying content.
At the same time, even where special labeling under Article 50 is not required, the image must not mislead the buyer about the real appearance or properties of the product.
Why a “created with AI” caption may not be enough
Article 50 covers more than visual disclosure to a human. For certain synthetic content, there must be a technical way to determine its AI origin.
In practice, a company needs to check:
- whether the AI tool in use adds machine-readable labeling;
- whether it is preserved after editing, exporting, and optimizing the file;
- whether the CMS or marketplace strips the necessary metadata;
- whether the user is informed about AI content clearly enough;
- whether the disclosure method meets the needs of people with disabilities.
A caption on the page may be clear to a visitor but still fail to meet the technical part of the requirement. Conversely, hidden metadata does not always provide sufficient transparency for the user.
Who is responsible for AI content
One of the main practical problems is that several parties are often involved in creating and publishing the material.
For example:
- a provider supplies a generative AI system;
- an agency uses it to create an image or text;
- a company employee reviews and edits the result;
- an online store publishes the material for customers.
You cannot assume that the AI tool developer or an external agency has automatically fulfilled all obligations. The company publishing the content must understand its origin, how it was processed, and the rules that apply to it.
Key risks for business
Handling AI content incorrectly creates not only regulatory risks, but also commercial ones.
- Lack of transparency. The customer does not understand that they are talking to AI or viewing synthetic material.
- Loss of technical labeling. Metadata disappears after image processing or upload to the CMS.
- Misleading the buyer. The generated image does not match the real product.
- Unclear responsibility. The contract with the contractor does not specify who reviews and labels the content.
- No inventory of AI tools. Different departments use services that legal and management teams do not know about.
- Inaccessible disclosure. Labeling exists, but it is hard to notice or cannot be read with assistive technologies.
Why automated checks are not enough
An automated scanner can find individual technical signals, but it cannot reconstruct the full content lifecycle or determine each company’s role on its own.
During a review, you need to establish:
- which AI systems employees and contractors actually use;
- what content they create or modify;
- where and for which audience it is published;
- how substantial the AI involvement was;
- who performs human review and bears editorial responsibility;
- which technical markings remain in the final file;
- how the disclosure appears to a real user.
That is why a complete assessment requires a combination of technical checks, process analysis, and expert interpretation of the requirements.
How to prepare for Article 50 requirements
Preparation should start not with adding the same caption to every piece of content, but with an inventory of AI use.
- List your AI tools. Include services used by marketing, sales, support, HR, design, and external contractors.
- Identify types of AI content. Record where texts, images, audio, video, chatbots, and virtual characters are used.
- Split scenarios by risk level. Distinguish ancillary editing from generation and substantial modification of materials.
- Check technical labeling. Make sure the necessary markings survive every processing stage.
- Set up human review. Assign people responsible for accuracy, transparency, and publication of content.
- Update processes and contracts. Define the duties of employees, agencies, and providers.
Check your business readiness for the AI Act
DigiWay Media audits the use of artificial intelligence: we analyze the website, content, AI tools, and internal processes, identify potential risks, and prepare a practical action plan.
As a result, you will receive:
- a map of the AI systems and scenarios in use;
- a list of materials that need further review;
- an assessment of current labeling and user disclosure;
- a list of priority changes before the requirements start to apply;
- recommendations for employees and external contractors.
FAQ
Do all AI photos need to be labeled?
The main criterion for visible labeling by the publishing company is whether the image falls under the definition of a deepfake: whether the generated or substantially modified content appears to be an authentic image of an existing person, object, place, organization, or event.
Do you need to label an AI background on a product photo?
Such an image may require labeling if the generated background creates the impression of a real place or situation. Allegro expressly treats a real product on an AI-generated background as an example of material that a seller needs to assess.
Who is responsible for an image from a supplier?
If a company publishes the material in its own offer, it needs to check its origin and determine whether labeling is required. A contract with a supplier or agency may allocate responsibility between the parties, but it does not remove obligations toward the consumer and the regulator.
Do AI texts on a blog need to be labeled?
Not all of them. Article 50 specifically regulates AI texts published to inform the public on matters of public interest. An exception may apply where there is substantive human review, editorial control, and a person or organization that bears editorial responsibility.
Is putting an AI icon on the image enough?
Not always. Labeling must be clear, distinguishable, timely, and accessible. In addition, the business needs to review other AI scenarios and must not strip machine-readable information added by the AI system provider.
Do the requirements already apply?
Yes. The main transparency requirements of Article 50 apply from 2 August 2026. A limited transitional period until 2 December 2026 covers only technical labeling by providers of certain AI systems placed on the market before 2 August.
