The EU AI Act in 2026: Which Transparency Rules Matter Now?

Further parts of the European Union’s AI Act have applied since 2 August 2026. Transparency surrounding certain AI systems and synthetic media is a particular focus. Businesses using chatbots, AI voices or generated images now need to distinguish more carefully between the disclosures envisaged for different uses and technical signals that serve a different purpose from visible labelling. These are EU rules, not requirements that automatically apply in every country.

What changed on 2 August

On 2 August 2026, further central provisions of the EU AI Act became applicable. Particularly visible are transparency rules for certain AI systems and artificially generated or altered content. The European Commission explained the forthcoming enforcement on 31 July. This includes situations in which people need to recognise that they are interacting with an AI system rather than a person.

Importantly, not every conceivable use of AI carries the same labelling obligation. Using a writing assistant to draft an internal outline for a blog article differs from publishing a convincingly realistic audio recording of a real person. The specific categories and exceptions in the regulation are decisive.

Distinguish chatbots, deepfakes and technical markings

For certain interactive systems, transparency towards users means informing them that they are communicating with AI. The regulation provides particular disclosure obligations for deepfakes. Separately, it includes requirements for machine-readable markings of generated or edited content, which vary by role and use case.

These concepts are often mixed together in public discussion. A visible note saying “AI-generated illustration” differs from an invisible technical watermark. Nor does a Content Credential automatically satisfy every legal obligation. Businesses should therefore examine technical, editorial and legal measures separately.

What marketing teams should review now

Create an inventory of the AI tools you use and their applications. Are there chatbots on the website? Are synthetic voices used in advertising? Are real people simulated realistically? Who publishes content, and who first checks whether readers need additional information? Document these decisions in a traceable way.

An internal checklist for each content type is useful. An advertising video may raise different issues from a decorative illustration. Transitional and special provisions also exist and require careful reading. The current regulation and the competent authorities remain authoritative; this article is not individual legal advice.

Practical example: An AI voice in a product video

A company uses a synthetic voice to narrate a product video. The marketing team should first establish the type of content involved and whether it deceptively imitates a real person. It then checks what transparency information is required and how to present that information clearly within the user experience.

The team should also consider the voice’s usage licence, rights in the source material and its own brand communication. A properly documented approval makes later questions easier to answer. A generic “AI” sticker without context is no substitute.

Conclusion

Since August 2026, AI transparency has become a concrete organisational issue for many businesses operating within the EU rules’ scope. Distinguishing the different obligations and establishing appropriate approval processes brings clarity. My advice: start with actual use cases rather than general headlines. Seek qualified legal advice when obligations are unclear.

Sources and further reading

The following primary sources and official documentation provide further detail. Always check the current position on legal matters and topics that change over time.

• digital-strategy.ec.europa.eu
• ai-act-service-desk.ec.europa.eu
• eur-lex.europa.eu

This article provides general information for the German and European context and does not replace legal advice. Other jurisdictions may follow different rules.