Polish Companies Struggle to Interpret AI Content Labeling Rules

A month after the AI Act's transparency rules for AI-generated content took effect, companies in Poland still don't know exactly what they're required to do. Lawyers put it bluntly: some are labeling nearly every piece of content just in case, others assume the obligation doesn't apply to them at all, and both groups could be wrong.

What the Rule Actually Says

Article 50 of the AI Act distinguishes four categories of situations covered by the transparency obligation. The first concerns systems that interact directly with people, such as chatbots or voice assistants, whose providers must design them so users know they're talking to a machine. The second category covers synthetic content, meaning text, audio, images or video generated or modified by AI. The third includes emotion recognition and biometric categorization systems, and the fourth covers deepfakes and text published on matters of public interest.

The key mistake many companies make is assuming that every piece of text edited with AI needs a note like "content generated by artificial intelligence." Lawyers stress that the rule works differently: the obligation to label the tool's actual output generally falls on the provider of the AI system, not on the company using it. A publisher mainly has to disclose to its audience deepfakes and texts on matters of public interest that haven't undergone genuine human verification.

Where the Confusion Comes From

The problem started with oversimplified media coverage that, over the past few weeks, suggested every piece of content made with AI assistance needs a visible label. Some companies overreacted and began adding AI notices to almost every publication, even where artificial intelligence served merely as a supporting tool rather than a source of content. Other businesses drew the opposite conclusion and decided the new rules don't apply to them at all.

The paradox is that both groups could be wrong - Jarosław Chałas, managing partner, Chałas i Wspólnicy Law Firm

Based on the guidance, lawyers say the key question every business should ask is whether the obligation arises in its activities at all. The answer depends on where the entity sits in the AI Act chain, meaning whether it's a provider or a deployer of the system, what kind of system it uses, and exactly what it publishes.

Examples From Practice

The uncertainty plays out in everyday professional situations. A marketer who polishes a finished text with an AI tool doesn't automatically have to label anything, as long as the final material went through their editorial review. It's a different story for a graphic designer using an image generator for inspiration, a creator running an avatar-hosted online training session, an influencer using a synthetically generated voice, or an agency producing promotional materials in the form of a deepfake depicting a real person.

For multimedia content, the obligations are broader than for text alone. If a video, audio or graphic material depicts an existing person, place, object or event in a way an audience could mistake for authentic, a notice that it's AI-generated must appear at the first point of contact with that material. Exceptions cover works that are clearly artistic, satirical or fictional in nature, provided the audience is properly informed of that, as well as materials used by law enforcement acting within the law.

What It Means for Polish Companies

For Polish businesses, the problem is a practical one, because the lack of clarity creates risk of two kinds of mistakes: needlessly damaging a brand's image by over-labeling material that doesn't require it, or genuinely exposing the company to administrative liability by publishing content on matters of public interest without the required disclosure. Marketing firms, newsrooms, creative agencies and online creators are the groups most exposed to both mistakes, since their work most often combines AI tools with content reaching a wide audience.

The situation is further complicated by the absence of a fully established national AI Act oversight body in Poland, something lawyers and industry organizations have been flagging for months. Without clear, Polish-language guidance from a supervisory authority, companies have to rely on interpretations from law firms and explanations from the European Commission, which don't always reach practitioners in an easily understandable form.

What Comes Next

The European Commission is trying to limit the interpretive chaos by publishing explanations of Article 50 along with a free set of icons for labeling AI content, whose use remains voluntary, even though the underlying obligation to inform audiences is not. More sector-specific guidance can be expected in the coming months, especially for industries hit hardest by the uncertainty, such as media, marketing and online education. Until then, companies in Poland will have to assess the risk themselves, most often with help from lawyers specializing in AI regulation.