European writers decline to co-sign EU AI transparency code over loopholes

European writers and literary translators have said they will not co-sign the Code of Practice on Transparency of AI-Generated Content promoted by the European AI Office, arguing that its exemptions could allow AI-generated or manipulated content to reach the public without adequate disclosure. In a joint statement issued on International Artists’ Day, the European Writers’ Council and the European Council of Literary Translators’ Associations said the code, promoted by the European AI Office, did not deliver meaningful transparency. The groups said responsibility could be circumvented at different stages of distribution, from AI providers and users through publishers, platforms and distributors to readers. They called for disclosure to remain with material throughout that chain, backed by mandatory machine-detectable marking and human-readable labels. Detailed information about generated or manipulated material should also remain accessible to readers. “Transparency must be persistent,” the organisations said. Their objection focuses partly on terms including “human review,” “editorial control” and “AI-assisted fiction,” which they said were overly broad or undefined. The groups argued that human review cannot turn AI-generated work into human-created or human-written work, and therefore should not remove disclosure duties. Article 50 of the EU AI Act includes an exemption for content that has undergone human review or editorial control. The law requires providers of systems that generate synthetic text, audio, images or video to make outputs machine-readable and detectable as artificial or manipulated, as far as technically feasible. But the marking duty does not apply to standard editing assistance or systems that do not substantially alter a user’s input or its meaning. Article 50 also requires disclosure when AI-generated or manipulated text is published to inform the public on matters of public interest. That requirement does not apply where the material has undergone human review or editorial control and a person or organisation takes editorial responsibility for publication. The Commission is to encourage and facilitate Union-level codes of practice on detecting, marking and labelling such content. For the book sector, EWC and CEATL said a rule centred on marking an output at the moment it is generated or manipulated was insufficient. Print books, e-books and audiobooks require different technical approaches, they said, while metadata and watermarks can be removed or altered as content is copied, transformed and redistributed. Services intended to remove or evade watermarks are already appearing, they added. The organisations also questioned whether watermarking can work without public means of verification. They said Anthropic and OpenAI do not freely make available the software necessary to identify their source watermarks, and called for system-relevant selection and detection tools to be available to everyone as open-source software. Translations were a particular focus. A literary translation whose first draft was wholly or partly made with generative AI should be treated as AI-generated, the groups said, and identified on the title page, copyright page and in metadata. Someone hired to revise such a translation should be described as an editor rather than a translator. EWC and CEATL warned that a broad exemption for AI-generated or manipulated entertainment could let products imitating novels and poems evade labels. They said AI-generated or unlicensed machine-translated books had been distributed commercially at scale since 2023, and estimated that about half of new e-book titles were presumed to be machine-generated. The organisations cited a 2025 Eurobarometer finding that more than eight in 10 Europeans prefer human-created works. They argued that presenting generated works as human-created or human-translated could result in improper remuneration and compromise support schemes for human creators. Collective management societies, literary and translation funds, and state arts-funding systems were already confronting fraud linked to AI-generated products, they said. Copyright and authors’ rights, the groups added, were essential to transparency but absent from the code. The EWC, which says it represents more than 260,000 professional authors through 55 associations in 34 countries, has previously argued that machine-generated books are competing directly with authors. In September, it said translator commissions in five larger book markets had fallen by one-third over five years, while illustrator commissions had declined by one-fourth.