European Writers and Translators Refuse to Sign the EU's AI Transparency Code

Two of Europe's biggest author groups say they will not co-sign the EU's code on labelling AI-generated content. In a joint statement published October 25, timed to International Artists' Day, the European Writers' Council (EWC) and the European Council of Literary Translators' Associations (CEATL) said the Code of Practice on Transparency of AI-Generated Content "falls short of what is needed to achieve meaningful transparency."

Together the two groups say they represent around 272,000 writers and literary translators across Europe. In its final version, they wrote, the code "risks becoming technically useless for the book sector," because too many exemptions let responsibilities slip along the distribution chain.

What the code is

The code is a voluntary framework, drawn up by independent experts in a process run with the EU's AI Office, to help AI providers and deployers meet the transparency rules in Article 50 of the EU AI Act. The European Commission published the final version on June 10, 2026, and the Article 50 obligations, including clear labels on deepfakes and on AI-generated text published on matters of public interest, have applied since August 2, 2026.

The writers' main objections

Transparency has to follow the book. The groups argue that marking can't stop at the moment an AI system produces text. It should follow the content from the AI provider to the user, then to publishers, platforms, distributors and the reader, with both machine-detectable marking and human-readable labels at every step.

Watermarks aren't enough. Technical marks can be stripped or altered when content is copied or transformed, the statement says, and a market is already growing for services that remove them. The groups add that unless AI companies such as Anthropic and OpenAI make detection software freely available, no one outside those companies can check for watermarks in generated text. They called that "a mockery of compliance" and asked for open-source detection tools.

The exemptions are too broad. The statement objects to carve-outs built on undefined ideas such as "human review," "editorial control" and "AI-assisted fiction." A human review, it says, "cannot magically transform AI-generated content into human-created or human-written works." The groups also criticize a broad exemption for AI-generated entertainment products, including content made to look like novels or poems.

Translations should be labelled. EWC and CEATL say a literary translation should count as AI-generated if any of its first draft was produced with generative AI. In that case, they want it stated on the title page, the copyright page and in the metadata, and they say the person hired to revise such a draft should be credited as an editor, not a translator.

The money behind it

The groups tie the issue to pay. Human-written and human-translated work carries copyright payments, secondary rights and access to public support schemes. If AI products can be presented as human work, the statement warns, that money gets diverted, and funds and collecting societies are already facing fraud. Authors who don't use AI often end up having to prove their work is human, it says, when the burden should fall on AI-generated products. The statement also cites a 2025 Eurobarometer survey finding that more than eight in ten Europeans prefer works made by humans.

Why it matters for writers

Signing the code is voluntary, but it's the EU's main practical yardstick for how AI content gets labelled. If the book-sector carve-outs stay as they are, readers buying ebooks or audiobooks in Europe may have no reliable way to tell whether a title was written, or translated, by a person. For independent authors who sell into the EU, this is the clearest statement yet of what Europe's writer organizations want labelled, and where they plan to keep pushing.

Sources

• European Writers' Council and CEATL, October 25, 2026: Exemptions should not swallow the rule: joint statement on the Transparency Guidelines and the Code of Practice under Article 50 of the AI Act