Publishing and research specialists have questioned the practical effectiveness of the European Union's AI Act during a discussion at the 2026 Frankfurt Book Fair, highlighting concerns about copyright enforcement, transparency and the reliability of AI-assisted scientific publishing.
The October 7 panel brought together representatives of publishing, academia and information standards to examine how artificial intelligence is changing the production, verification and distribution of scientific knowledge.
Elizabeth Crossick, head of EU government affairs and global AI policy at RELX, discussed the legislation's requirements for general-purpose AI models, including policies for compliance with European copyright law and the disclosure of information about training content.
Although the AI Act establishes obligations and financial penalties, Crossick questioned whether its implementation provides sufficient incentives for AI developers to comply. She also pointed to the complexity of the accompanying regulatory measures and guidance.
Gerhard Lauer, professor of book studies at Mainz University, raised a different concern: the changing relationship between information and its original sources.
He explained that AI systems process material without necessarily preserving distinctions between books, journals and other forms of publication. This can make it more difficult to establish where information originated and who should receive attribution.
Wolf-Tilo Balke of TU Braunschweig addressed the increasing volume of scientific submissions, including AI-generated research material. He argued that traditional peer-review processes face growing difficulties in maintaining reliable assessment as submission volumes rise.
Todd Carpenter, executive director of the US National Information Standards Organization (NISO), highlighted the TRACE initiative, a collaborative effort to develop methods for preserving provenance and attribution as AI systems retrieve and process scholarly content.
The initiative brings together organisations working in scholarly publishing, information standards and research infrastructure. Its projects include developing metadata models and investigating how AI systems access, use and attribute published material.
The panellists differed over the appropriate regulatory response.
Carpenter supported flexible approaches that could accommodate future technological changes, suggesting that commercial demand for reliable and verifiable AI systems might also encourage improvements.
Crossick argued that stronger preventive requirements and clearer liability provisions would be preferable to relying primarily on legal action after copyrighted material had already been used.
She also referred to the Frankfurt Appeal, an initiative through which publishing organisations have urged European policymakers to improve transparency and the practical enforcement of existing copyright protections.
The discussion did not announce new legislation or changes to the EU AI Act. Rather, it examined the challenges of applying existing requirements to rapidly developing technologies.
For publishers and researchers, the central questions concern how copyrighted works are used, whether information can be traced to its original source, and how the reliability of scholarly communication can be maintained as artificial intelligence becomes more widespread.