AI Content in TYPO3: Labelling Needs Accountability

Editorial teams use AI to translate, summarise and rewrite text, while images and videos are increasingly generated or edited with AI. This makes two questions more important: How was this content created, and who took responsibility for publishing it?

The first question has come into focus with generative AI. The second is much older. In a professional editorial environment, knowing who last edited a piece of content is not enough. What matters is who reviewed this exact version, approved it and cleared it for publication. The transparency obligations under Article 50 of the EU AI Act, which apply from 2 August 2026, highlight something content management systems should have supported for a long time: traceable origins and clear editorial accountability. With AI Label, we are providing a ready-to-use solution for TYPO3.

The AI Act distinguishes between providers of AI systems and the companies or organisations that use them professionally. Providers of generative AI systems must ensure that AI-generated or manipulated content can be identified in a machine-readable format. Additional disclosure obligations apply to professional users, particularly for deepfakes and certain AI-generated or manipulated texts on matters of public interest.

This does not mean that every piece of content created with the help of AI automatically needs a visible label. For text, one relevant factor is whether it has undergone editorial review and whether a person or organisation takes responsibility for its publication. Human sign-off therefore matters alongside the technical origin of the content.

The European Commission explains these distinctions in its guidelines on transparency obligations. The authoritative source remains Article 50 of the EU AI Act.

At first glance, the task appears simple: add a field for “AI-generated” and “AI-edited” in the backend and display a label in the frontend. In practice, AI transparency affects the data model, interfaces, editorial workflows and rendering. An image can be used in many different places in TYPO3, content elements and project-specific content types use different templates, and a growing number of extensions and external systems create content directly within the CMS.

A reliable solution must therefore record how the content was created, which system provided that information, who reviewed it and who approved this exact version. It must then make the relevant information reliably available in the frontend.

AI Label distinguishes between two forms of AI involvement:

• AI-generated: The content was created entirely using artificial intelligence.
• AI-edited: Existing content was changed using artificial intelligence.

Content without AI involvement continues to work as before and does not receive an additional label. We have deliberately chosen not to introduce a separate deepfake category. The term describes a legally relevant use case, not the technical origin of the content. A deepfake was still either generated or edited using AI. TYPO3 therefore does not have to assess whether a piece of content legally qualifies as a deepfake. AI Label records what is known about its origin and makes that information available to the backend and frontend.

Other extensions can set the AI status directly. If for example the Ext:T3ppygenerates or edits content using AI in the future, the editorial team does not have to add this information afterwards. Any system that generates or edits content using AI should be able to provide information about its origin at the same time. AI Label does not attempt to detect AI-generated content automatically or make legal decisions about individual cases—it records known information and passes it on consistently.

The origin of a piece of content is only one part of the transparency we need. Human accountability is at least as important. TYPO3 can already record who created or last edited a record, but that does not automatically mean that this person reviewed the content and approved it for publication.

This is why AI Label adds a sign-off. It records which person took responsibility for the published version. The sign-off deliberately refers to a specific version: If the content is changed after approval, the previous sign-off must not simply apply to the new version. It must remain clear what was reviewed and who was responsible for the content that was actually published.

This is not an AI-specific problem. In an enterprise workflow, content creation, editing, review and publication may all be handled by different people. Every professional publishing process therefore needs a clear answer to one question:

Who decided that this exact content, in this exact version, was ready to be published?

The AI Act merely makes it more obvious how often this information is still recorded outside the CMS, buried in comments or handled through informal agreements.

TYPO3 projects look different and use a wide range of content types. AI Label therefore does not impose a fixed frontend design. Integrators decide which icon to use, how to phrase the notice and where it should appear. AI Label provides the meaning—the project determines the presentation.

The label must be clearly associated with the relevant content, understandable and visible without additional interaction. A notice that only appears on hover, in metadata or in the source code is not enough. An icon alone is also not sufficiently clear from our perspective. The output should therefore include understandable, localised wording, such as:

This content was generated entirely using artificial intelligence.

The label must also be implemented accessibly. An image’s alternative text remains unaffected: it continues to describe the content or function of the image, not how it was created.

AI Label solves this challenge today as a TYPO3 extension. In the long term, information about content origins and approval should not remain something every project has to implement separately. A professional CMS should be able to record how content was created, which version was reviewed and who took responsibility for its publication. This applies to AI-generated text just as much as it does to press releases, product information, medical content or university pages.

TYPO3 therefore needs a shared foundation in the Core: a standardised data model for content origins, interfaces for extensions and external systems, and a sign-off linked to the exact version approved for publication. The frontend presentation should remain part of the individual project, but the underlying information and approval process should be standardised.

The AI Act does not create an entirely new requirement for content management systems. AI simply makes visible what professional workflows have needed for a long time.