European Commission Unveils First Draft Icons for Labeling AI Content

The European Commission has shown what labeling of AI-generated content could look like in practice. The published draft code of practice includes four sample icons meant to help audiences distinguish AI-generated material from content created by humans.

The document explains the obligations set out in Article 50 of the EU's AI regulation, known as the AI Act. The provision imposes specific information obligations toward content recipients on providers of generative AI systems as well as on the entities deploying them. The code is meant to serve as a practical guide to meeting those obligations in everyday practice, without waiting for additional guidance or case law.

Four Proposed Icons

In Annex I of the draft, the Commission presented four sample graphic solutions. One, prepared using the free version of ChatGPT from December 2025, takes the form of a minimalist two-color "AI" label. A more elaborate concept came from studio Artifact, which introduces three content categories: H for fully human-made material, AI for content fully generated by artificial intelligence, and AI-H for hybrid content, with an indication of the percentage of AI involvement in a given piece.

The examples themselves are illustrative and do not prejudge the final shape of the regulation. The Commission says it ultimately wants to introduce a single, uniform, interactive EU icon that would inform recipients whether a given piece of content was wholly or partly generated or modified by AI. Clicking the icon would give access to detailed information about the material's origin, including the tool that created it.

Obligations for Providers and Deployers

The draft code splits obligations between two groups of entities. Providers of generative AI systems are to apply multi-layered labeling mechanisms combining markers visible to users with machine-readable solutions, provide tools for verifying the origin of content, and ensure that labels cannot easily be removed or manipulated.

Companies deploying AI systems, essentially any entity publishing content created with such tools, will have to label deepfakes from the very first moment a recipient encounters the material, apply the uniform EU icon, and label public content that has been generated or substantially altered by AI. This also applies to artistic, satirical, or fictional content, which has raised concerns among some creators about how labeling might affect the reception of a work.

no single active labeling technique is sufficient on its own to meet all the legal requirements

A Technical Challenge Without a Ready Solution

The draft's authors openly acknowledge that no single technology today can fully solve the problem of labeling AI content. This means companies' compliance will be assessed based on the entirety of their compliance processes, documentation, and testing of solutions, rather than simply the deployment of one specific technical tool. That is a practical pointer for legal and technology departments already preparing for the August deadline.

For companies, this means rolling out training, human oversight procedures for AI-generated content, and extensive compliance processes before the regulation takes effect. The lack of clear technical standards complicates implementation, however, since companies still do not know exactly which labeling format the Commission will ultimately approve.

What This Means for Poland

In Poland, work is proceeding in parallel on a law on AI systems, which would establish the Komisja Rozwoju i Bezpieczeństwa Sztucznej Inteligencji (Commission for the Development and Safety of Artificial Intelligence) as the central oversight body, with representatives from UODO (Poland's data protection authority) and UOKiK (Poland's competition and consumer protection authority). Polish companies publishing AI-assisted content, from media outlets to online stores using generative product descriptions, will have to comply with EU requirements regardless of the pace of work on domestic implementing legislation.

Consultations on the draft code run until January 23, 2026, so the final shape of the icons and labeling mechanisms may still change. A second draft of the document is expected in March, with a final version in June 2026, shortly before the labeling obligation takes effect on August 2. Until then, companies operating in the EU market will need to track successive versions of the code to implement compliant solutions in time.