The European Publishers Council strongly objects to the third draft of the AI Code of Practice on the basis that it fundamentally fails to meet the objectives of the EU AI Act and undermines existing EU copyright law. While the AI Act was designed to support responsible AI development in Europe which respects and supports its cultural and creative sectors, the current draft of the Code moves in the opposite direction - to the detriment of innovators, creators and the broader public interest.
Rather than providing a clear and meaningful compliance framework, the Code introduces legal uncertainty, misinterprets key legal provisions, and establishes an unacceptably weak standard for credible compliance by General-Purpose AI (GPAI) providers. It reduces copyright compliance to mere “reasonable efforts” and narrowly focuses on web crawling, ignoring other widespread data collection methods for AI development that are equally subject to copyright law, and fall within the rubric of the AI Act. This blind spot creates significant risks of infringement but also legal exposure for AI developers and downstream providers.
Critically, the draft dismisses the reservation of rights by publishers and other rightsholders by: continuing to focus solely on robots.txt, an unfit for purpose voluntary protocol which is frequently ignored or circumvented; refusing to recognise natural language reservations, and failing to require transparency in how rights are respected. AI bots that bypassed robots.txt grew by over 40% between Q3 to Q4 2024. Even when blocked, bots continue to access sites through unidentified user agents which was roughly equivalent to identified AI bot scraping in Q4 2024.
It also unjustifiably limits transparency under the guise of trade secrets, even though the unauthorised use of third-party content for AI training cannot be considered confidential business information. Rightsholders have a legitimate and overriding interest in understanding whether, how, and to what extent their content has been used, and in assessing the lawfulness of such use.
The absence of meaningful obligations to obtain authorisation from rightsholders—including through collective management organisations—represents a missed opportunity to facilitate responsible AI innovation and instead drives rightsholders to enforce their rights through technical means and litigation. This does not serve an open internet, nor does it lead us a step towards a collaborative future where AI innovation and human creativity can both flourish.
The EPC has engaged constructively throughout this process, but the drafters have ignored reasonable substantive input from across the creative sectors. This third draft undermines the very purpose of the AI Act and weakens the position of rightsholders in the face of powerful AI companies who refuse to engage in constructive negotiations. As such, the EPC must reject the draft in its current form. No Code would be better than one that legitimises non-compliance, misleads the market and dilutes the already weak incentives to foster greater collaboration amongst right holders and AI developers.
EPC has also co-signed the Joint statement by a coalition of authors, performers and other rightsholders active across the EU’s cultural and creative sectors.
More information on the opacity, pace of growth and extent of scraping by AI crawlers can be found the recent report from Tollbit which provides an AI User Agent Index updated quarterly that tracks overall trends in the scraping habits of the AI industry.