As of August 2, 2026, Article 50 of the AI Act (Regulation (EU) 2024/1689) is no longer a plan for the future, but a binding obligation backed by actual oversight.
The transparency rules apply to two groups: providers of AI systems and deployers who use them in their business operations. If your website has a chatbot, you generate product descriptions using a language model, or you publish AI-generated images, at least one of these obligations applies to you. The European Commission has published guidelines under Article 50 and a Code of Good Practice, so the excuse “we had no way of knowing” no longer holds water.
Съдържание
Who is the supplier and who is the implementer?
This distinction defines exactly what obligations you have. A provider is a party that develops an AI system or markets it under its own name, while a deployer is any organization that uses the system in a professional context.
For a typical WooCommerce store, the role is almost always that of an implementer. The store does not train its own models but uses off-the-shelf services: a chatbot plugin, the OpenAI or Anthropic API for descriptions, and an image generator for banners. The responsibility for machine-readable markers falls primarily on the model provider. However, the obligation to label deepfakes or certain AI-generated texts remains yours.
The Four Obligations Under Article 50
The list is short, but it covers almost every generative use.
- Chatbots and other interactive systems must clearly state that the other party is an AI, not a human
- AI-generated or manipulated content (audio, images, video, text) must include a machine-readable marker so that it can be identified as synthetic
- Deepfake content is clearly labeled for the user
- AI-generated texts that inform the public on matters of public interest must be labeled as such, unless they have been edited by a human editor who assumes editorial responsibility.
The exceptions are narrow. Standard processing such as cropping, color correction, or noise reduction does not turn a photo into AI-generated content, and assistive uses such as spell-checking do not make the text synthetic. The Commission’s guidelines, updated on July 20, 2026, provide specific examples of what falls within the scope and what remains outside it.
What Is a Machine-Readable Marker?
Here, the law builds on existing technologies. The marker could be metadata compliant with the C2PA standard, an invisible watermark embedded in the pixels, or a statistical fingerprint within the text itself. Google, for example, embeds the SynthID watermark in images from Imagen and in text from Gemini, while OpenAI adds C2PA metadata to files from DALL-E. The requirement is that the solutions be effective, interoperable, and sustainable, to the extent that this is technically feasible.
The problem is that resizing, compression, or even a simple screenshot often destroy the metadata. That is why the Commission is talking about a combination of techniques, rather than a single universal method.
The Code of Good Practice and the EU Icons
On June 10, 2026, the Commission published the final Code of Practice on Transparency of AI-Generated Content. The document has two sections: rules for providers (labeling and detection of synthetic content) and rules for implementers (labeling deepfakes and AI-generated text). Signing the code is voluntary, but the obligations under Article 50 are legally binding and apply to everyone. As of July 31, 2026, the code has been signed by over 180 organizations, and the Commission and the AI Council have confirmed that it is an adequate means of demonstrating compliance.
Any entity that does not join will have to demonstrate to the supervisory authorities that its own measures are equivalent.
The EU has also published a set of icons to label AI content, which implementers can use as-is. This unified visual system solves a real problem: users see the same label in Germany, Bulgaria, and Spain, rather than each website having to come up with its own badge.
Checklist for WordPress and WooCommerce Websites
The practical steps are summarized in a short list.
- Check to see if your chatbot plugin clearly displays a message indicating that the conversation is with an AI; most SaaS solutions added this feature in 2026.
- If you publish AI-generated news or analysis on topics of public interest without editorial intervention, please include a visible label
- Ask the model provider what markers it embeds in the images and text, and keep the response as evidence in case of an audit
- Describe the use of AI in the website’s terms and conditions; ready-made legal templates for online stores are available at obshti-uslovia.com
- AI-generated product descriptions are not a “matter of public interest” and do not require a visible label, as long as they do not mislead the consumer
If you’re just starting to plan your online store, incorporate these practices from day one, rather than adding them under pressure during testing. The cost is close to zero when the chatbot and the generated designs are tailored to meet requirements as early as the prototype stage.
EU Icons
The icons come in four versions: black, white, black with 50% transparency, and white with 50% transparency. You can download ZIP files containing all the icons in all versions in SVG and PNG formats.
The icons underwent user testing, and the results influenced their design. It is noteworthy that performance improved across all metrics when the main icon was accompanied by a text label (e.g., “modified”).
| Icon | When is it used? | Examples |
Main Icon | When artificial intelligence has been used to create fake content (images, audio, video) or published text, or when a personalized text label or interactive overlay has been added. | A deepfake video with the text label “voices generated with,” followed by the main icon |
Generated entirely by artificial intelligence | When all fake content (images, audio, video) or text is generated entirely by artificial intelligence without any human input or human editorial oversight (except for prompts) | Deepfake videos featuring politicians or fictional events, generated entirely by artificial intelligence; Music or art entirely composed by artificial intelligence;* News summaries generated by artificial intelligence. |
Partially modified using artificial intelligence | When it existed previously, the human-generated content was partially modified using artificial intelligence, turning it into deepfake text or text on topics of public interest. | A person’s face in an authentic photo was replaced with that of a politician using artificial intelligence. Authentic photos of an empty apartment were generated using artificial intelligence. |
*May be used with limited disclosure requirements for artistic, creative, or satirical works
The use of these EU icons is optional, but the labeling requirements under Article 50 of the Artificial Intelligence Act are not. The use of these icons alone does not establish compliance with the law. Implementers remain responsible for ensuring that any disclosure complies with the requirements of Article 50 of the Artificial Intelligence Act. Signatories to the Code of Practice on the Marking and Labeling of AI-Generated Content must duly implement the measures set forth therein.
Who monitors compliance?
Supervision is shared among national market surveillance authorities, the Commission’s AI Office (for systems under its jurisdiction), and the European Data Protection Supervisor when it comes to EU institutions. Any user or competitor may file a complaint through the Commission’s AI Act Complaints Tool, including in Bulgarian. In Bulgaria, the national framework—including penalties and specific supervisory authorities—is still being finalized as of mid-2026, but the regulation is directly applicable and does not require the adoption of local legislation.
Learn more about the fines under the AI Act.
This topic is evolving rapidly, and it’s worth keeping an eye on the other AI-related posts on the blog, where you’ll also find technical analyses of tagging technologies.
Frequently Asked Questions
-
Should AI-generated product descriptions in an online store be labeled?
No. Visible labeling is required for deepfakes and for AI-generated text on topics of public interest that are not subject to editorial oversight. Machine-readable labeling is the responsibility of the model provider, while misleading descriptions fall under consumer rights.
-
Does the chatbot on my website need to say that it’s AI?
Yes, as of August 2, 2026, interactive AI systems will inform users that they are conversing with a machine, unless this is already obvious from the context.
-
What is a machine-readable AI content marker?
Metadata compliant with the C2PA standard, an invisible watermark such as SynthID, or a statistical fingerprint embedded in the text, which allow software to identify the content as generated or manipulated by AI.
-
Is signing the Transparency Code mandatory?
No, the code is voluntary. However, the obligations under Article 50 are legally binding on everyone, and those who have not signed it must demonstrate compliance with equivalent measures of their own to the supervisory authorities.
