EWC and CEATL refuse to sign EU AI transparency code

The European Writers' Council and CEATL announced they will not co-sign the European Commission's Code of Practice under AI Act Article 50, saying its final version is, in their words, "technically useless for the book sector" because too many exemptions let responsibility be circumvented along the distribution chain, according to the EWC's joint statement.

The two organisations, representing writers and literary translators across Europe, object to exemptions built on undefined terms such as "human review," "editorial control," and "AI-assisted fiction," which they say can let AI-generated or AI-manipulated content reach the public without disclosure. They also flag watermarking as unreliable, noting that identification tools are not freely available from AI companies named in the statement, Anthropic and OpenAI, which they say renders the method unusable by anyone outside those companies.

The statement cites the 2025 Special Eurobarometer on Europeans' attitudes towards culture, which found over 8 in 10 Europeans prefer human-created works over AI-generated content. It also states that, since 2023, machine-translated books have been distributed commercially at scale, with about 50% of new e-book titles presumably machine-generated.

Here is the part worth sitting with: the EWC and CEATL aren't arguing that the Code does nothing. They're arguing it does just enough to look like protection while leaving the actual loopholes open. That's a sharper complaint than "needs more work" — it's "this will be used against you."

Look at the translation rule they flag specifically. If the first draft of a translation is fully or partially produced by generative AI, the statement says it should be disclosed on the title page, copyright page, and in the metadata — and the person who revises that AI draft should be credited as editor, not translator. That's a clear, specific rule. The problem the two groups are raising isn't that rule itself — it's that terms like "human review" elsewhere in the Code aren't defined with the same precision, which they say leaves room for AI-generated content to clear the bar without the label ever showing up.

The watermarking fight matters to you too, even indirectly. The statement says machine-detectable marks can be stripped or altered once content is copied or distributed, and that an entire market is emerging to do exactly that. If marking is the only enforcement tool and it can be removed downstream, the "transparency" promise on paper doesn't travel with the book.

What to actually do this week: if you're a working translator, check your own contracts and metadata practices now against the standard the statement describes — first-draft AI involvement disclosed on title page, copyright page, and metadata, with the human reviser credited as editor. Don't wait for the Code to settle that for you. And if you're a member of a writers' or translators' body in Europe, this is the week to ask your own organisation where it stands on co-signing — silence here gets read as consent.