EU Rules on Labeling AI-Generated Content Take Effect August 2

Starting August 2, 2026, Article 50 of the European Union's AI Act comes into force across the entire bloc. Companies, freelancers and public institutions will be required to clearly disclose when someone is talking to a chatbot, and to label images, audio and video generated or substantially altered by artificial intelligence.

What exactly is changing

Article 50 of the regulation imposes two main obligations. The first applies to providers of AI systems, who must design chatbots, agents and avatars so that users immediately know they are interacting with a machine rather than a human. The second obligation falls on those deploying AI and concerns labeling synthetic content: images, video and audio that look or sound like real people, places, products or events, even if no one intended to mislead anyone.

The European Commission's guidelines define a deepfake more broadly than one might expect. The definition covers not only depictions of real people, but any material that a viewer could reasonably mistake for authentic, including a product photo in an advertisement, an apartment visualization with generated furniture, or a touched-up photograph with realistically added elements.

Who the rules cover

The labeling requirement is not limited to large tech companies. Under the European Commission's guidelines, it applies to any natural or legal person who regularly derives economic benefit from AI-based activity, meaning it also covers freelancers, online creators, journalists and doctors running private practices. The only exemption is for purely private, non-professional use, such as generating a holiday card for one's family.

The definition covers a much broader range of content than initially assumed, and applies to objects that simply look realistic, not only to existing people or things - Tomasz Zalewski, lawyer commenting on the new guidelines

Exemptions from the requirement

The rules include several exemptions. Content used for detecting and prosecuting crimes under the law does not need to be labeled. Artistic, satirical and fictional works may be labeled in a less conspicuous way, for instance in the closing credits or in a platform's description, as long as this does not interfere with the enjoyment of the work itself. Texts verified by a human, for which the editorial team takes full editorial responsibility, are also exempt.

The situation is different for advertising and commercial materials, where the label must be immediately visible to the audience. Lawyers still have doubts about cases where only a single element in an otherwise human-made piece comes from an AI generator, since interpretations on this point still vary.

Penalties and oversight in Poland

Violations of the transparency obligations carry administrative fines of up to 15 million euros or up to 3 percent of a company's global annual turnover for the previous year, whichever is higher. The regulation requires that penalties be effective, proportionate and dissuasive, meaning national supervisory authorities will have latitude in setting fine amounts in individual cases.

In Poland, oversight of compliance will fall to the Commission for the Development and Security of Artificial Intelligence (Komisja Rozwoju i Bezpieczeństwa Sztucznej Inteligencji), established under the law implementing the AI Act, which the president signed on July 24, 2026. Lawyers point out that the obligations take effect regardless of whether the national supervisory authority is fully constituted by August 2.

What it means for Polish businesses

For small and medium-sized businesses and freelancers, this means reviewing marketing materials, website content and customer communications to identify where image, audio or text generators have been used. In practice this affects, for example, online stores using generated product photos, agencies producing ads with a synthetic voiceover, or companies deploying customer service chatbots without clearly disclosing that there is no human on the other end.

Providers of systems placed on the market before August 2, 2026 have been given extra time to comply with the requirement to add technical, machine-readable markers to content, which only takes effect from December 2, 2026. The remaining transparency obligations, including disclosing chatbot interactions and labeling deepfakes, apply from August with no transition period.