The EU's AI Act now requires labeling synthetic text

The writer's problem

A publisher using an AI narration or translation tool inside the European Union needs to know which of its own obligations under the EU's AI Act, and which of its vendor's obligations, are actually in force, since the regulation phases in duties over several years rather than all at once. The European Commission's own regulatory framework page lays out that schedule, including transparency obligations that reached full application on 2 August 2026.

What the documents show

The Commission states that providers of generative AI 'must ensure AI-created content is identifiable' and that certain content, deepfakes and text generated to inform the public, requires clear, visible labelling. The page states the Commission 'starts enforcing AI Act rules and new transparency requirements on 2 August,' when the AI Office and national authorities become responsible for supervision and enforcement of those duties. This labelling requirement is distinct from, and later than, a separate set of obligations that already bound providers of general-purpose AI models. The Commission's own guidelines on GPAI obligations state such providers must maintain a policy to comply with Union copyright law and must make publicly available 'a sufficiently detailed summary of the content used for training' their models, duties the guidance describes as already in application before the 2026 transparency deadline took effect. The Commission has also published a voluntary Code of Practice on marking and labelling AI-generated content, offered as guidance rather than a binding rule in itself.

The editorial choice

An editor advising an author or small press on an EU-facing AI tool should separate two questions the Act treats separately: whether the tool's provider has published a training-data summary and copyright policy, a GPAI-provider duty, and whether AI-generated output the press itself distributes needs a visible label, the transparency duty that reached full application in August 2026. This is an editorial reading of the Commission's own framing, not a restatement of the full statute, which covers many categories of AI system beyond writing tools.

What stays with the author

The Commission's pages describe what the regulation requires of providers and, more narrowly, deployers of AI systems; they do not describe how a specific writing or narration tool has classified itself, which a publisher would need to confirm with that vendor directly. Nor do they resolve whether a given use falls within the Act's territorial scope for a press based outside the EU whose books are still sold there, a case-specific question the framework page does not answer.

• Has the AI tool's provider published the training-data summary and copyright policy the GPAI guidelines describe?
• Does content the press is distributing fall into a labelled category, such as a deepfake or AI-generated text presented as informing the public?
• Does the Act apply to this press's specific EU-facing distribution, given its scope provisions?

The obligations described here apply unevenly across the range of tools a writer might use, and a specific product's compliance status should be confirmed with its provider rather than assumed from the regulation's general timeline.