Regulation

AI-generated text

EU AI content marking grace period approaches December deadline

A grace period for certain AI systems already available in the European Union is due to end on December 2, 2026, according to EY.

EU AI content marking grace period approaches December deadline

The European Union's AI regulation aims to make texts, images, audio and video generated or substantially modified by AI identifiable. Information about a file's origin can be embedded in several ways, including metadata, digital watermarks or other machine-readable markers.

The regulation does not mandate one specific technology. Instead, the chosen method should reliably help identify AI-generated content in practice.

Deadline approaches for existing systems

The requirements have applied to newly marketed systems since August 2, 2026. Tools already available in the European Union before that date received a four-month grace period, which expires on December 2, 2026.

According to EY, the change may affect more than large international AI platforms. Hungarian businesses may also fall within its scope if they use or sell software that generates content with AI under their own name. This can include companies that incorporate or offer another developer's technology under their own brand.

Specific obligations can also depend on how a company uses a particular solution. The same business may be considered a user in one activity and a provider in another. Zala Mihály, the partner responsible for EY's IT and Technology Consulting services, said companies should review before the December deadline which technologies they offer, which they use only for their own purposes and what happens to the content produced with them.

Markers must survive processing

Compliance involves more than simply adding a technical marker: companies should also check whether origin information remains intact after content is downloaded, compressed, converted or uploaded to another system. Resizing an image, reformatting a video or uploading a file to a platform can remove such data.

A machine-readable marker is not the same as a label visible to users. The December deadline concerns primarily the former and does not mean every piece of AI-generated content must carry a prominent warning.

In some cases, however, the organization publishing material must also clearly inform users. This can apply when publishing deepfakes: artificially generated or modified images, audio or video that appear authentic or realistic. Visible labeling is not automatic for AI-assisted text either. Relevant factors include the purpose of publication, the role of AI, human review and whether a person or organization accepts editorial responsibility.