Mandatory AI Content Labeling Takes Effect, But Companies Still Don't Know How to Comply

As of August 2, 2026, every business, online creator, and marketing agency in the European Union must label content generated by artificial intelligence. Article 50 of the EU AI Act took effect on schedule, but law firms and industry media agree that the provision does not specify exactly what the label should look like, and the line between a permissible photo edit and a deepfake requiring a label still sparks interpretive disputes.

What exactly must be labeled

The provision broadly covers deepfakes, meaning images, video, or audio generated or modified by AI that could be mistaken by a viewer for authentic material. The definition doesn't require the material to depict a specific, real person or place - it's enough that it resembles something that could plausibly exist. A generated plush toy in an ad flyer or an office rendering in a job listing both meet this condition, even though neither depicts any real object.

Content created solely in a private capacity is exempt, such as a digital holiday card sent to friends. A drawing of a dragon also doesn't need a label, since dragons don't exist in reality, so no one could mistake it for a real photo. The trouble starts with borderline material, sharpness correction, a color change, or minor retouching, where the law draws no clear line between an acceptable edit and a fabrication requiring a label.

Who the rule covers

The obligation applies to anyone working professionally with AI, regardless of the scale of their operation. In practice, that means a car repair shop owner posting an AI-generated 'before and after' photo, an influencer receiving a brand barter deal for posts featuring AI material, a marketing agency producing ads, and an online store whose chatbot must be disclosed as a machine unless that's obvious from the context of the conversation.

Responsibility for labeling rests with the creator of the content, not the party that commissions it. If an ad agency produces AI-generated material for a brand, the agency is responsible for labeling it, not the brand that publishes it under its own name. This division of responsibility has become a source of disputes in contracts between clients and contractors.

Why clarity is lacking

The law sets criteria for labeling - it must be clear, distinguishable, and visible from the recipient's first contact with the content - but doesn't provide ready-made label text. The European Commission has offered sample wording, such as 'AI Generated' or 'Photo modified by AI,' along with optional icons marked 'AI GENERATED' and 'AI MODIFIED,' but using them remains voluntary, leaving room for inconsistent practice across the market.

Another source of uncertainty is judging things from the recipient's perspective. The same animation that's obviously computer graphics to an adult might look real to a four-year-old, and the rule requires taking the target audience into account. Regulations on synthetic audio are even less precise than those for images - it isn't entirely clear whether AI-generated instrumental music requires a label in every case.

The European Commission recommends particular caution with content aimed at children and older adults - for them, the line between real and generated material can be harder to recognize.

How companies should implement it

Lawyers recommend four steps for companies: inventorying all AI-generated content, visibly labeling advertising and commercial materials, explicitly disclosing chatbots as machines, and organizing contracts with subcontractors so it's clear who's responsible for labeling any given piece of material. Where the label goes depends on the format - for images, it should be a persistent label in the corner of the frame; for video, a notice at the start or an overlay for the full duration; and for audio, a spoken disclosure at the start of the recording.

Artistic and satirical material is treated more leniently - the label can be moved to, say, a film's closing credits instead of appearing in the opening frame. However, the law doesn't precisely define where satirical work ends and commercial material subject to the full labeling requirement begins.

What this means for Poland

In Poland, the Sejm (the lower house of Poland's parliament) passed the law on artificial intelligence systems on June 11, 2026, by a vote of 421 to 3, and the Commission for the Development and Security of Artificial Intelligence formally launched under it on July 3. However, the commission won't gain full enforcement powers, including the right to impose fines, until October 28, 2026, meaning that for nearly three months the legal obligation is already in force while the national enforcement body still lacks the full tools to police it.

Labeling content as AI-generated doesn't exempt anyone from other laws - GDPR (RODO) provisions on personal data processing, criminal code provisions on identity theft, image rights protections, or fair competition rules. An 'it's AI' label isn't a pass to use someone's face or voice without consent, which lawyers stress is a common misconception among businesses preparing to comply with the rule.