Transparency obligations for AI: New rules from August

Since August 2, 2026, companies have been required to clearly label AI-generated content and communication via AI systems. An expert explains what to consider.

Since August 2, 2026, new transparency obligations have been in effect for companies that use artificial intelligence (AI) or publish AI-generated content. Providers and operators must now inform people that they are communicating via AI or viewing artificially generated content. Lawyer Klaus-Dieter Franzen answers the most important questions.

What is the background to the new regulations?

The European transparency obligations under Article 50 of the AI ​​Regulation (AI Act) apply to providers and operators of AI systems. Providers develop AI systems or market them under their own name. Operators deploy AI systems under their own responsibility. If an employee uses an AI system on behalf of and under the control of their company, the company generally remains the operator. The nature and scope of the obligations depend on this distribution of roles.

What applies then, and why?

People must be able to recognize when they are interacting with an AI system . For chatbots and virtual assistants, it is not sufficient to hide the information in the legal notice or terms and conditions. The notice can only be omitted if the AI ​​interaction is clearly evident . However, this exception must be interpreted narrowly. The information must also be accessible and clearly presented. Age-appropriate information is required for children.

How should AI-generated content be labelled?

From August 2, 2026, providers must mark image, audio, video, and text content from generative AI systems in a machine-readable way. This can be achieved, for example, via metadata or digital watermarks.

You might also be interested in:

Operations management

Construction site travel as working time: When do you get paid?

Operations management

Optician wins against data privacy challenger

Panorama

"Open doors with the mouse" on October 3rd

For AI systems that came onto the market before this deadline, a transitional period applies until December 2, 2026. Content created before August 2, 2026 is generally not subject to the labeling requirement.

What are deepfakes and what rules apply to them?

Anyone publishing realistic-looking AI-generated images, videos , or audio must clearly indicate their artificial origin at the first point of contact. A mere technical marking is insufficient. The indication must be recognizable without special technical aids (see Article 50(4) of the AI ​​Regulation) . This also applies, in principle, to completely artificially generated persons, places, or events as soon as they appear authentic. No intent to deceive is required!

Caution is particularly advised for photorealistic advertising and product depictions : A reference to artistic or satirical exceptions does not always protect!

Only unrealistic or cartoonish content is excluded.

Which AI-generated texts need a hint?

The regulation also requires a label for AI-generated texts that provide information on topics of public interest . Only a comprehensive professional review by a competent person can exempt them from this labeling requirement, provided that person assumes editorial responsibility. A mere spelling or formatting check is insufficient.

What are the rules regarding emotion recognition and biometric categorization?

Anyone using AI systems for emotion recognition or biometric categorization must actively inform those affected – both in real-time evaluation and in subsequent analysis. In addition, the requirements of the General Data Protection Regulation (GDPR) and the rights of works councils remain in effect.

What are typical mistakes?

Technical marking and visible disclosure must not be confused . Machine-readable marking is usually the responsibility of the provider, while visible disclosures are the responsibility of the companies distributing the content. Companies should not establish a superficial approval process. Those relying on editorial reviews need a clearly documented process with expert oversight and unambiguous accountability.

Many companies mistakenly believe that all AI obligations have been postponed. But: Article 50 of the AI ​​Regulation comes into force on August 2, 2026 ! Anyone who violates it risks fines of up to €15 million or up to three percent of their global annual turnover.

Practical tips

"Companies should promptly identify which AI systems they use and what content they generate," advises lawyer Franzen. "Customer service, HR, marketing, and corporate communications are particularly relevant. Clear instructions must be provided for chatbots. A labeling process is required for AI-generated images, videos, and synthetic voices. Contracts with external providers should stipulate whether and how machine-readable tags are provided. For publications, the company should define who reviews, approves, and is legally responsible for the content. This review process should be transparently documented . Furthermore, for company-wide applications, the GDPR and any potential co-determination rights of the works council must be observed."

Source: VDAA

DHB now also available digitally! Simply click here and register for the digital German Crafts Journal (DHB)!