
Insights · Compliance
AI Act transparency obligations: What companies should review now
The transition period for certain generative AI systems ends on 2 December 2026. These review points help companies prepare.
- Published
- 29 October 2026
- Reading time
- 5 min read
2 December 2026 is unlikely to feature prominently in most companies’ calendars. Yet the date is relevant to the use of generative AI: it marks the end of a transition period for certain technical marking obligations under the European AI Act.
Companies that use AI to create text, images or other content should therefore review which requirements apply in each case.
The challenge often lies in everyday processes: an AI tool in marketing, an integrated office feature or a publication process with unclear responsibilities. Without an overview of these applications, it is difficult to reliably assess when labelling is required.
What has applied since August 2026
The transparency obligations under Article 50 of the AI Act have generally applied since 2 August 2026.
The provision distinguishes between providers and deployers of certain AI systems. Providers must, for example, ensure that synthetic content is technically identifiable as AI-generated. Companies using such systems, by contrast, may be required to disclose to the public that certain published content has been artificially generated.
Further obligations concern, among other things, AI systems that interact directly with people, as well as applications for emotion recognition or biometric categorisation.
There is no blanket labelling obligation for every use of AI. The company’s role, the system used, and the nature and use of the content are decisive.
Which transition period ends in December
For certain generative AI systems already placed on the market before 2 August 2026, a transition period ends on 2 December 2026. It concerns providers’ technical marking and detection obligations under Article 50(2).
According to the German Federal Network Agency (Bundesnetzagentur), AI-generated content that had already been created and published before 2 August 2026 does not need to be labelled retrospectively. For later publications, however, a disclosure obligation may apply.
The December deadline is therefore a useful opportunity to review the company’s own use of AI and its existing publication processes.
Four review points for companies
1. Record AI systems and providers
Companies should first establish an overview of their AI applications. This also includes features integrated into marketing, office, HR or customer platforms.
Key information includes the provider, area of use and whether a system generates or modifies content or merely provides assistance. Companies should also establish whether they act as a provider or a deployer.
An AI inventory of this kind forms the basis for a targeted review of regulatory requirements.
2. Distinguish between types of content and publication channels
Not every AI-generated piece of content requires visible labelling. Two categories are particularly relevant:
Deepfakes: Where AI-generated or manipulated image, audio or video content is published and could falsely appear authentic, its artificial creation must generally be disclosed.
Texts concerning matters of public interest: Where AI-generated text is published to inform the public about such matters, a disclosure obligation may also apply.
An internal draft, an ordinary product description and a publicly accessible specialist article may therefore require different assessments. Companies should map their typical content types and publication channels accordingly.
3. Organise labelling and editorial control
Providers’ technical markings and deployers’ disclosure obligations serve different purposes. A machine-readable marking does not automatically replace a required disclosure to the public.
Article 50(4) contains an important exception for AI-generated text: the disclosure obligation does not apply where the content has undergone human review or editorial control and a natural or legal person holds editorial responsibility for its publication.
A simple spelling correction or formal approval will generally not be sufficient. An actual review of the substance is required.
A specialist article created with AI assistance, reviewed by a subject-matter expert and published under clear editorial responsibility therefore does not necessarily have to be labelled as AI-generated.
4. Assign responsibilities
Internally, it should be clear who assesses AI applications, reviews outputs and approves publications. It should also be clear when Legal or Compliance needs to be involved.
Not every publication requires a comprehensive individual assessment. Standardised assessments, approval rules and templates may be sufficient for recurring use cases.
Clear responsibilities and practical workflows are decisive.
What companies should document now
An elaborate parallel organisation is not required. Often, it is enough to make targeted additions to existing AI policies and approval processes.
A concise overview should show which AI systems are used, which types of content are affected, which transparency obligations or exceptions apply, and who is responsible for review and publication.
Companies should also record what provider information on technical marking is available and how required disclosures or editorial controls are implemented.
This makes it possible to meet regulatory requirements in a traceable way without creating unnecessary bureaucracy.
Conclusion: Transparency obligations need clear processes
The AI Act does not require companies to label every AI-generated text or image. What matters is identifying relevant use cases and implementing the legal requirements in a targeted way.
Although the transition period ending on 2 December 2026 directly concerns only certain provider obligations, it offers a timely opportunity to review the company’s own AI governance.
Companies that understand the systems they use, their publication channels and their responsibilities can usually implement transparency obligations with manageable effort.
Support with AI governance
Clarius Legal helps companies assess their AI applications from a legal perspective and implement the AI Act in practice. This includes reviewing regulatory obligations, developing suitable AI policies, and designing responsibilities and approval processes.
Our approach combines legal certainty with practical solutions for everyday business.
Do you have questions about this topic?
We can assess what the development means for your organisation and advise you directly.


