News
Allegro Tells Sellers to Label AI Images Ahead of EU AI Act's August Deadline
Starting August 2, 2026, online sellers across the EU will have to tell buyers when product photos or videos were generated or substantially altered by artificial intelligence. Allegro has already emailed sellers about the change and launched a beta tool for labeling such images.
Contents
Allegro has emailed sellers warning that in two weeks, new AI Act rules on AI-generated content will take effect across the European Union. Starting August 2, 2026, anyone listing a product with a photo or video generated or substantially modified by AI will have to clearly disclose that to buyers.
What exactly must be labeled
The provision applies chiefly to content classified as deepfakes, meaning images, videos, and audio recordings generated or significantly altered by AI in a way that creates an impression of authenticity. For an online seller, that means virtual models showing off clothing, AI-generated product photo shoots, synthetic video footage, or voiceovers reading out a listing's description.
The European Commission has clarified that content doesn't need to depict a real person or an actual place to fall under the labeling requirement, it just needs to look realistic. A tomato generated from scratch for a grocery store ad or a photorealistic rendering of an apartment with synthetic furniture also require a label. Images of a dragon or other things that obviously don't exist are exempt.
Exception for text and editorial content
Article 50(4) carves out an exception for text content that has undergone editorial review and for which a specific person or company takes responsibility, but that exception applies only to text, not images, video, or audio. Standard product descriptions, buying guides, or content articles produced with the help of language models remain outside the main scope of the new obligations, even if they were written by ChatGPT or another model.
The obligation applies from August 2, 2026, unchanged - Marta Adranowska
Significant modifications that create a false impression of authenticity should be treated as deepfakes. EU guidance indicates that what matters is the viewer's impression, not the creator's intent - Tomasz Zalewski
Who is liable and what penalties apply
Responsibility for labeling falls on whoever creates the AI content, not the platform that distributes it. If an advertising agency produces a product deepfake for a brand, the agency, not the brand, bears the legal liability, though the brand still risks unfair competition claims. Penalties for violations reach 15 million euros or 3 percent of annual global turnover, and in Poland they are to be enforced by the Komisja Rozwoju i Bezpieczeństwa Sztucznej Inteligencji (KRiBSI, the Polish AI Development and Security Commission), even if it is formally established only after the rules take effect.
The label must be clear and visible no later than the moment a user first interacts with the content. Standards vary depending on the type of material - for artistic, satirical, or fictional content, a mention in the credits or video description suffices, while ads or commercial content need a label that's immediately obvious.
What it means for Polish sellers
Allegro is the first major platform in Poland to start actually implementing the rule, expanding its listing data structure with an aiCoCreatedContent field and adding the ability to declare images as AI-generated directly in the API. For sellers using product photo generators or virtual models, that means updating their integrations before August 2 to avoid disruptions in managing listings.
A practical problem remains with metadata and digital signatures such as Content Credentials, which the AI Act envisions as a complement to visible labels - some systems strip them out when the same content is published simultaneously to a store, an ad campaign, and social media. That means automatic annotation in a single system alone doesn't guarantee compliance everywhere the content ends up.
What's next
The August 2 deadline covers only part of the AI Act's provisions, obligations for providers of the most powerful models and high-risk systems are phased in over subsequent years under the law's timetable, partly pushed back by the Omnibus VII package. For now, online sellers and advertising content creators will be the first to feel the new obligation in their day-to-day work, and Allegro may not be the last platform to start sending similar warnings in the coming weeks.
