EU AI Act – What online sellers should know now

The EU AI Act is raising a lot of questions among online sellers: First, the good news – for most online sellers, there’s no reason to panic. In fact, a large portion of the new obligations primarily affects providers and developers of AI systems – for example, companies that develop their own AI applications or use high-risk AI. Those who use AI merely as a tool in their daily work – for example, for product descriptions, translations, marketing texts or customer communication – must still observe some basic requirements.

 

 

1. Employees should be proficient in using AI

 

The AI ​​Act requires companies to provide their employees with appropriate training in the use of artificial intelligence. Anyone who works with AI should not only be aware of its possibilities, but also – and especially – its risks and limitations. Employees must receive training specifically for this. It is recommended that companies provide their employees with guidelines in this regard.

 

 

2. Typical violations that online sellers should avoid in connection with AI :

 

GDPR violations:

• Copying customer names, addresses or e-mail addresses into ChatGPT.
• Pasting complaints with complete customer data into an AI tool.
• Entering personal data or confidential business information.

Trademark/copyright infringements:

• AI creates a product image that closely resembles a well-known brand logo or a protected character.
• AI copies lengthy passages of text from external sources.
• AI generates images in the style of a specific artist or uses protected trademarks.

Competition law violations:

• AI adds incorrect product features (e.g. “waterproof” even though the product is only splash-proof).
• AI ​​claims certifications or seals of approval that do not exist.
• AI makes up guarantees or technical data.
• AI creates impermissible health-related advertising claims.
• AI ​​advertises with unique selling points such as “market leader”, “test winner” or “no. 1”, without these being verifiable.

 

 

3. The responsibility always remains with the company

 

One of the most significant sources of error is AI chatbots! AI should never communicate with customers without prior review; rather, the company must ensure in advance that the AI used for customer communication is not permitted to invent anything that has not been expressly approved by the company as information. Otherwise, there is a risk of costly warning letters, which some courts have already upheld.

AI can significantly simplify daily work. What it can’t do is take responsibility. Anyone who publishes content remains responsible for it, even if it was created by AI. Product descriptions, advertising claims, legal notices or health-related statements should therefore always be reviewed by an employee.

 

 

4. Transparency is becoming more important

 

The AI Act also establishes transparency requirements for certain AI applications. For example, users must be able to recognise whether they are communicating with an AI system. A simple message such as “You are chatting with our AI assistant” should generally suffice. Labelling requirements also apply to certain images, videos or audio content generated entirely by AI. A seller is not required to label every AI-generated product image as “AI-generated”. Similarly, a product description written with AI support does not need to be labelled as such. The transparency requirements of the AI Act apply only when the image or video gives the impression that a real person or event is being authentically depicted, even though it was in fact artificially generated or manipulated.

 

 

5. AI images on the Kaufland marketplaces: What sellers need to know

 

As a general rule: If you, as a seller, upload product images or videos created with AI, this content must be properly labelled as such. However, on the Kaufland marketplaces you don’t have to worry about it yourself: The multi-marketplace technology behind the marketplaces handles and manages this entire labelling process completely automatically in the background.

The system reads the invisible metadata that each AI-generated image provides and automatically adds a visible, legally compliant AI label to the images when they are uploaded to the platform.

For this automation to work, you only need to ensure that the metadata contained in the images remains unchanged so that it is sent to the system!

 

 

6. What sellers need to do now: Protect the metadata!

 

To ensure that automatic labelling works and the system can read the data, the image metadata must not get lost. Therefore, you must observe the following rules when processing images:

• No screenshots: Always download images directly from the AI ​​tool. Screenshots do not contain this metadata.
• Caution when exporting: When using image editing programs (e.g. Photoshop), make sure to actively preserve the metadata when saving.
• Avoid format changes: Some tools only save labels in certain formats (e.g. PNG). However, these can be lost during conversion (e.g. to JPEG).
• Do not send via a messaging service: Do not send images via WhatsApp, Teams or social media, as these platforms remove metadata during upload.
• No manipulation: Deliberately removing the label is illegal under the AI ​​Act.

6.1 Checking the files for valid AI labelling

You can check for free whether a file has a valid AI label before uploading it:

1. Drag the file into the C2PA Viewer inspection tool.
2. The check takes place directly in the browser – the file is not uploaded to external servers.
3. In the results, look for the DigitalSourceType entry with the value trainedAlgorithmicMedia: If this entry is present, the AI label has been retained.

Do you have any questions or need help? Kaufland Global Marketplace Seller Support is happy to assist you and support you in selling on the Kaufland marketplaces.