Legal status as of: October 13, 2026, Marta Lampart, , Attorney-at-Law
Effective August 2, 2026, Article 50 of the AI Act—which sets forth transparency requirements regarding artificial intelligence—will take effect. This does not mean that every post or graphic created with the help of AI must be labeled. The obligation depends on who you are (the provider of the tool or the person or company using it) and what you are publishing: a realistic image, video, or audio (deepfake), text on public matters, or a chatbot.
W tym artykule wyjaśniam po kolei, kogo dotyczą nowe przepisy, kiedy oznaczenie jest konieczne, a kiedy nie i jak zrobić to w praktyce, żeby nie narazić się na karę.
Since when has AI content labeling been required?
The most important date is August 2, 2026 As of that date, Article 50 of the AI Act—that is, Regulation (EU) 2024/1689—takes effect. The provisions are directly applicable in all EU countries and do not require a Polish law to take effect.
There has been quite a bit of confusion in recent months because, in July 2026, a regulation known as the Digital Omnibus on AI went into effect, which pushed back some of the deadlines under the AI Act. So it’s worth sorting out the dates:
- August 2, 2026 – all obligations under Article 50 take effect, including the obligation for individuals and companies using AI to disclose deepfakes and certain texts.
- December 2, 2026 – by that date, providers of AI tools that were on the market before August 2, 2026, have until then to implement technical (automated) labeling of generated content. This is the only transition period provided for in Article 50.
- October 28, 2026 – In Poland, the provisions of the Act on Artificial Intelligence Systems — which cover, among other things, oversight, proceedings, and penalties — are now in effect. Oversight is exercised by the Commission for the Development and Safety of Artificial Intelligence (KRiBSI).
Myth: “The AI Act has been postponed.” The Digital Omnibus primarily deferred the obligations regarding high-risk systems (to 2027 and 2028). The transparency obligations under Article 50 were not deferred, except for a technical aspect on the part of providers.
Who is subject to this requirement? The provider or the processor
The AI Act divides responsibilities between two roles. Each role is assessed separately for each tool and its intended use.
AI System Provider This is the entity that develops an AI system and makes it available under its own name (e.g., a company that develops an image generator or a chatbot). The provider must:
- design a system so that people know they are interacting with an AI (chatbots, voice assistants, avatars),
- ensure that the generated images, video, audio, and text are labeled in a machine-readable format and can be identified as synthetic.
An entity that uses an AI system This includes any person or company that uses an AI system in its business operations: an entrepreneur, an online store, an agency, a graphic designer, a photographer, or an influencer. The user must:
- disclose that a realistic image, video, or audio clip (deepfake) was generated or manipulated by AI,
- disclose that a text published to inform the public about matters of public interest was generated by AI (with the exception I describe below),
- inform people if it uses an emotion recognition system or biometric categorization.
A common question is: Who is liable when an AI-generated ad was created by an agency? As a general rule, the liability rests with whoever uses AI to create the content. If realistic graphics were generated by an agency or a graphic designer, then they should be responsible for labeling them, not the brand that commissioned the work (although opinions on this are divided, and each case must be analyzed individually). However, the brand is not entirely in the clear: an unlabeled deepfake in an ad could expose it to allegations of misleading consumers or unfair competition. That’s why it’s worth addressing this in the contract. Keep in mind that there are European Commission guidelines on this matter, but ultimately, the Court of Justice will determine what the correct solution is.
A social media platform is not required to tag content on your behalf. The user who created the content is responsible for tagging it.
Do I have to label every post or image created with AI?
No. The AI Act does not impose a general requirement to label everything created with the help of artificial intelligence. The requirement applies to specific categories of content.
Exception for private activities. The regulations do not apply to individuals who use AI as part of their personal and non-professional activities. According to the European Commission’s guidelines, both conditions must be met simultaneously. An AI-generated holiday card sent to family does not require a disclosure. But if you maintain a profile through which you regularly earn money (including through barter and collaborations), your activity is considered professional. Conversely, publicly criticizing a politician using a deepfake is not considered a purely personal activity, even if you do not earn money from it.
The standard edition is not the same as synthetic content. Adjusting the sharpness or colors, or removing a minor blemish with an AI tool, does not usually create a deepfake. However, a far-reaching alteration that can make something that never existed appear authentic does.
| Situation | Should I tag it? |
|---|---|
| Color and Focus Correction for Product Photos using an AI Tool | Usually not (but it depends) |
| Realistic AI-generated photos of a product or model for advertising | Yes |
| Realistic interior visualization with furniture added by AI | Yes |
| A recording featuring a synthetic voice or a cloned human voice | Yes |
| A fairy-tale illustration of a dragon or a clearly unrealistic character | Usually not (it's not a deepfake) |
| An Instagram post about your offer written with the help of ChatGPT | Usually not (this isn't an article about matters of public interest) |
| An article about changes in the law generated by AI and published without fact-checking | Yes |
| An AI-generated holiday card sent to loved ones | No (personal and non-professional activities) |
The assessment always depends on the specific content. The guidelines emphasize that what matters is the impression made on the audience you’re addressing, not your intention. Particular caution should be exercised with content intended for children and seniors.
What is a deepfake as defined by the AI Act?
In everyday language, a deepfake is a fake recording of a well-known person. In the AI Act, the term has a much broader meaning: it refers to an image, video, or audio clip generated or manipulated by AI that resembles real people, objects, places, or events and could be mistaken for the real thing.
The European Commission’s guidelines from July 2026 showed that a deepfake does not have to depict a specific, real person. It is sufficient for the content to appear realistic. Therefore, a deepfake can include, for example, a generated image of a vegetable for a promotional flyer, a realistic visualization of an apartment, or a video featuring a synthetic voice. However, something that clearly does not exist—such as a dragon or an elephant driving a car — would not be considered a deepfake.
Artistic, satirical and fictional content they are not exempt from this obligation, but they are treated more leniently. If a deepfake is part of a clearly artistic, creative, satirical, or fictional work, it is sufficient to disclose the use of AI in a way that does not detract from the work’s impact—for example, in the end credits or in the video description. Advertising and business materials should be clearly and immediately labeled.
In practice, there are also controversial situations. For example, lawyers disagree on whether a travel video featuring instrumental music generated by AI needs to be labeled. If your situation is borderline, it’s safer to label it than to take a risk.
AI-Generated Text: When should it be labeled?
This requirement applies to texts published to inform the public about matters of public interest: news, analyses, commentary on current events, and articles on changes in the law or policy.
This designation is not required if the text has been actual verification of a person or editorial review, and a specific person or company bears editorial responsibility for it. Simply proofreading the text or casually clicking “I accept” is not enough.
A typical marketing text about your offer, a product description, or a sales post is usually not a text dealing with matters of public interest, so this obligation does not apply to it. Keep in mind, however, that separate rules apply under consumer law and the UOKiK’s recommendations regarding the labeling of advertisements.
A chatbot on your website or in your online store — what do you need to do?
If a website uses a chatbot, the person using it must be aware that they are interacting with an AI. This information should be provided no later than the first interaction, such as in the chat window. The exception is when this is obvious to an attentive and reasonably informed user.
Formally, this obligation falls on the system provider, but if you’re implementing a ready-made tool, check whether such a message actually appears and update your terms of service and privacy policy accordingly. A separate blog post about chatbots (GDPR, terms of service, liability for responses) will be published soon.
How do you properly label AI-generated content?
The AI Act don't impose a single labeling format. Instead, it specifies what the label must include:
- clear and precise – the recipient can't miss it,
- provided no later than at the time of the first contact in the body of the text, rather than hidden at the end of a long description,
- understandable – in simple language, in the audience’s language,
- available – including for people with disabilities (e.g., alternative text, voice narration in audio recordings).
The European Commission has released a set of EU icons to indicate the use of AI. You can use them, but you don’t have to. The icon itself does not guarantee compliance: what matters is whether it is visible, whether it appears at the right time, and whether the information remains visible after the content is downloaded or shared.
The Commission has also published a code of conduct regarding the labeling of AI-generated content. Adherence to the code is voluntary, but the obligations under Article 50 apply regardless.
Content published before August 2, 2026 According to the Commission’s guidelines, there is no need to retroactively label them, although the Commission recommends doing so whenever possible. If you reuse old content, such as in a new campaign, evaluate it according to the new rules.
What are the consequences of failing to provide the required labeling?
For violations of transparency obligations, the AI Act provides for penalties of up to 15 million euros or up to 3% of total annual global revenue, depending on which amount is higher. More lenient rules for determining penalties have been established for smaller companies.
In Poland, penalties will be imposed by the Commission for the Development and Safety of Artificial Intelligence (KRiBSI). The regulations on oversight and penalties take effect on October 28, 2026. The fact that the commission is still in the process of being established does not mean there will be impunity: the labeling requirement has been in effect since August 2, 2026.
Labeling AI-generated content is different from labeling an advertisement. If you publish a sponsored post featuring AI-generated imagery, you may need two labels: information about the advertisement (in accordance with UOKiK recommendations) and information about the use of AI.
Checklist for Creators and Entrepreneurs – Labeling AI-Generated Content
- List AI tools, which you and your team use, including AI features in graphics software, office suites, and video applications.
- Determine your role for each tool: Do you use a ready-made solution (end user) or do you offer it under your own brand (provider)?
- Check what kind of content you're posting: realistic images, video, audio, articles on public issues, and a chatbot on the website.
- Adopt a single labeling rule (where, with which phrase or icon, at what point) and use it consistently.
- Complete the contracts with agencies, graphic designers, and creators: who labels AI-generated content, who is responsible for any lack of labeling, and whether the labels should be retained upon publication.
- Implement an AI usage policy w firmie, nawet krótką i przeszkol osoby, które publikują treści.
- Document decisions: if you believe that a particular piece of content does not need to be flagged, please explain why.
FAQ: Frequently Asked Questions About AI Content Labeling
Do I need to label graphics created in Canva using AI?
It depends on the effect. If you’ve created a realistic image of a person, product, or place that a viewer might mistake for real, and you’re publishing it professionally, label it as such. Graphics that are clearly illustrative or abstract are usually not considered deepfakes.
Does “influencer” necessarily mean photos edited with AI?
Minor adjustments (lighting, colors, focus) usually do not require labeling. Changes to the appearance, silhouette, or background, or the addition of elements that were not originally present, may create a false impression of authenticity and require labeling.
Do I need to attribute text on my blog that was written using ChatGPT?
Only if it is a text reporting on matters of public interest and has not undergone actual verification by a person for whom someone bears editorial responsibility. A text that you fact-check and sign with your own name usually does not require a label.
Who is responsible when the AI-generated content was prepared by an agency?
The obligation to disclose the source of the content lies with whoever used AI to create it—typically the agency—but it also depends on who issued the instructions. However, the brand may also be liable on other grounds, such as misleading consumers, so it’s worth addressing this in the contract.
Does this requirement apply to content published before August 2, 2026?
According to the European Commission’s guidelines, there is no requirement for retroactive attribution, although the Commission does recommend it. Apply the new rules to every new publication, including when you reuse older material.
Is the hashtag #AI enough?
Most of the time, no. The label should be clear, distinct, and visible upon first encountering the content. A hashtag hidden among many others does not meet this criterion.
Summary – AI Content Labelling
Starting August 2, 2026, you won’t have to label everything created with the help of AI, but you must label realistic images, videos, and audio, as well as specific texts, and disclose the use of chatbots. Today, creators and companies that publish realistic advertising content — without realizing that, under the AI Act, it constitutes a deepfake — face the greatest risk.
If you want to quickly see which of your content needs to be tagged, check out my e-book „Oznacz albo nie?” – link to the e-book (in Polish) in the store. If you need an analysis of your situation or contract terms for agreements with agencies and creators, schedule a consultation or visit the website Artificial Intelligence.
___
About the Author Marta Lampart is a legal advisor and has been running her own law firm for many years. Her practice focuses, among other things, on intellectual property law and artificial intelligence law: she helps creators, influencers, and entrepreneurs protect their brands and content and use AI in compliance with the law. More about the author..
This article is for informational purposes only and does not constitute legal advice regarding any specific case.
