Major Danish publishing house Lindhardt og Ringhof has imposed a strict ban on the use of artificial intelligence tools by its staff and external contractors, Danish public broadcaster DR reports.
The policy was outlined in a directive sent to all outside collaborators including proofreaders, editors, manuscript readers, and literary consultants as well as internal editorial employees. Under the new guidelines, individuals handling materials for the publisher are prohibited from using AI systems for their work.
Protecting Copyright and Proprietary Work
The decision follows growing concern within the publishing industry over intellectual property, copyright exposure, and data privacy. According to the publisher’s letter, even seemingly administrative tasks such as uploading unpublished manuscripts or drafts to generate summaries, proof text, or check formatting can pose serious legal risks.
According to DR, the company in their letter warned that materials uploaded to commercial AI platforms can subsequently be copied, stored, and scraped for data mining or model retraining, regardless of whether tool providers claim user data remains private.
A Preemptive Move
Lindhardt og Ringhof’s literary director, Sune de Souza Schmidt-Madsen, explained for DR that the publisher had not caught external partners misusing generative AI, but felt immediate action was needed.
“It is only a matter of time before we run into this problem, and we want to be ahead of the curve”, Schmidt-Madsen added. 
Beyond copyright and security concerns, the publisher emphasized the importance of human judgment in literary work, telling partners that preserving the traditional craftsmanship of editors, translators, and readers remains central to the publisher’s standards and its duty of confidentiality toward authors.
Lindhardt og Ringhof’s ban, targets a major practical loophole in how European copyright law and the EU AI Act handle Text and Data Mining (TDM).
In July 2026, the European Commission published about a „Study to assess the feasibility of a registry of text and data mining opt-outs expressed by rightholders“.
The study finds that „current opt-out mechanisms are fragmented, inconsistently implemented and often ineffective. Source-based signals, such as robots.txt, lack granularity and persistency, while work-based metadata, such as TDMRep or IPTC, are frequently stripped or not reliably processed. As a result, rightholders struggle to exercise or verify reservations of rights, and AI developers lack a trusted reference point for demonstrating compliance“.
Under European law, training an AI model on existing work is treated as text and data mining, governed by the 2019 EU Copyright Directive (CDSM Directive) and reinforced by Article 53 of the EU AI Act.
