AI Content Labeling 2026: New Obligations for Businesses, Influencers, and Creators

Starting August 2, 2026, the EU AI Act enforces strict transparency obligations and fines of up to 15 million euros for anyone professionally publishing AI-generated content.

One Date That Impacts Millions of Ordinary Online Accounts

Up until now, obligations stemming from EU artificial intelligence law sounded like a problem for Silicon Valley corporations. That ends this coming Sunday. From August 2, 2026, anyone professionally publishing content created or altered by AI—from an influencer promoting cosmetics, to the owner of a sole-proprietorship workshop, to an advertising agency and a store with a chatbot—has a new obligation: they must clearly label that it is artificial intelligence. Ignoring this risks fines reaching 15 million euros. We explain in simple terms exactly what is changing, what a “deepfake” truly means under the new regulations, and how to prepare in a few minutes, with commentary from legal counsel Katarzyna Załęska.

AI Act—meaning the European Union artificial intelligence regulation—has been discussed for two years primarily in the context of technological giants. On August 2, 2026, that narrative changes. A section of the regulations enters into force that will for the first time realistically affect ordinary internet users and small entrepreneurs: the obligation of transparency. In practice, this means that content generated by artificial intelligence—photos, videos, audio, as well as chatbots—will in many situations need to be clearly labeled.

This is not another corporate procedure. This is a rule you will see scrolling through Instagram, watching an ad on Facebook, or writing to a “consultant” on a store’s website.

What the AI Act Actually Is in Three Sentences

The AI Act (EU Regulation 2024/1689) is the world’s first comprehensive law regulating artificial intelligence. It entered into force on August 1, 2024, but its provisions activate in stages over several years. The key principle: the greater the risk that a given AI application poses to people, the stricter the requirements—ranging from bans (e.g., so-called citizen social scoring) to full freedom (like spam filters or AI in games).

As legal counsel Katarzyna Załęska explains, the structure of this regulation resembles the already familiar GDPR: “The Artificial Intelligence Act has the rank of an EU regulation, and therefore, similarly to GDPR which has been in force for eight years already, the regulation applies directly, without the necessity of passing additional national regulations. (…) Instead of regulating ‘AI as such,’ the regulation grades obligations depending on what potentially negative impact a given application may have on citizens.”

For most of us, the most crucial layer of this pyramid is the transparency obligations from article 50. And these are precisely the ones that start applying on August 2.

What a Deepfake Means in Human Terms

Colloquially, a “deepfake” is associated with a swapped face or a falsified recording of a politician’s or celebrity’s speech. However, the definition from the AI Act is significantly broader. It is any image, video, or audio generated or altered by AI that resembles reality to such an extent that a recipient could consider it genuine.

The crucial word here is “resembles.” Katarzyna Załęska explains this with surprising examples: “The content does not have to depict a specific, real person or thing. It is enough that it looks like something that could exist in reality. Therefore, a deepfake can be a very realistically generated stuffed animal in an advertising flyer of a toy store or an office image in a job posting visualization—these are objects that generically exist in the world, only that particular specific copy does not exist.”

There is also a boundary of common sense. Not everything created by AI is a deepfake: “A drawing of a dragon does not require labeling because dragons do not exist—no one will mistake it for a photograph. However, it should be remembered that the assessment is always made through the prism of a specific group of recipients. A four-year-old confronted with a video about a unicorn or flying Santa Claus sleighs may consider it authentic.”

Who the AI Act Really Applies To

The provision speaks of a “deployer”—which sounds official, but in practice refers to anyone who uses AI professionally. And “professionally” is understood here very broadly: it is regular derivation of benefits from it—including minor ones, including in-kind ones.

Small business owners and sole proprietors fall under these rules regardless of scale or follower count if they generate AI content for business purposes. Influencers publishing AI videos promoting cosmetics via barter deals or brand packages are likewise acting professionally under the AI Act and must label their content.

Marketing agencies and graphic designers must note a crucial detail: the obligation rests on whoever creates the content, not on the client ordering it. If a professional designer generates an advertising deepfake for a business, the responsibility under Article 50(4) rests on the designer.

Exceptions and Technical Verification

You do not need labeling if you send birthday wishes with an AI-generated flower bouquet to a friend privately and non-professionally. Artistic creation and satire also have exceptions where labeling can be postponed, such as to the end credits of a movie.

Enhancing photo sharpness, colors, or removing a background using AI does not usually create a labeling obligation. Further-reaching interference, such as changing body proportions or adding elements, crosses into generating false impressions of authenticity. Voice cloning of an existing artist is a deepfake regardless of context, while synthetic voices undergo a recipient impression test.

Violating transparency obligations carries administrative fines of up to 15 million euros or 3% of total annual worldwide turnover, whichever is higher. In Poland, the competent body for imposing fines is the Commission for Artificial Intelligence Development and Safety (KRiBSI).