Labeling and Traceability of AI-Generated Material: What Do the New Requirements Entail?

In an interview with Joakim Karlén, we discuss new requirements for AI-generated material, primarily various types of labeling. The discussion stems from the new regulations that recently came into force as a result of the AI Act.

New Requirements

The EU AI Act and its new transparency requirements place higher demands on how AI-generated material must be handled and labeled. For many organizations, the new rules raise questions regarding what responsibility actually lies with them—whether as a developer, employer, or individual publisher.

The transparency rules in the EU AI Act have now entered into force, two years after the legislation's original adoption. This means that all organizations that develop or deploy AI systems need to comply with clear requirements for labeling and traceability.

Types of Labeling

Labeling of AI-generated content can broadly be divided into two categories:

Overt labeling: Clearly visible information for the viewer, such as a watermark on an image stating "AI-generated" or plain text in a document's metadata.

Technical (invisible) labeling: Mathematical or statistical shifts in the material, such as SynthID. It is not visible to the human eye at first glance, but can be identified by digital detection tools.

Who Bears the Responsibility?

It is important to distinguish where the responsibility lies in the chain. Creators of AI models (such as OpenAI, Anthropic, and Google) are required to build traceability and labeling directly into the technology. For companies that purchase or deploy these tools for their employees, the same general requirements for automatic labeling do not apply—with important exceptions.

When publishing deceptive material, such as deepfakes or completely unmoderated texts, an obligation arises to inform the recipient. If, on the other hand, text or images have been reviewed, edited, or processed by a human prior to publication, an explicit warning is generally not required.

As AI tools become a natural part of daily work—much like earlier technologies such as spell check—the need for major adjustments diminishes. The key for companies is to establish clear internal principles for communication and transparency. By proactively clarifying how AI is used in the organization, ambiguities are avoided and trust among customers and the public is maintained.

The full interview with Joakim can be found here.