The Frankfurt Book Fair opens its doors on Tuesday, October 6, but only to industry agents and professionals, a policy that has drawn criticism from the French delegation of rights directors and France Livre, a French publishing organization. Before the official opening, the Frankfurt Rights Meetings, a series of online conferences, allowed professionals from different markets to share best practices, with a consistent point regarding artificial intelligence (AI): it is necessary to be trained before using it. The first challenge, according to some, is not exposing the house and its authors to potential risks associated with AI. At S. Fischer Verlag, Elisa Diallo, the rights director, describes clear rules and internal training, but the house is "mainly in an experimental phase," without rushing. AI is used for administrative tasks such as contracts and royalty statements, not for translation, and no rights holders have yet asked for permission to use it. Regina Brooks, from Serendipity Literary Agency, summarizes the stance by stating that "one does not negotiate with what one refuses to understand." Limiting AI to the back office with safeguards, the agent uses tools like Storywise (management of spontaneous manuscripts), Royalty Reader (royalty statements), and Colaris (submissions and payments). Her condition is that the tools must be certified SOC 2, a security standard that ensures suppliers manage client data with a high level of protection and confidentiality, keeping data in a closed environment and not feeding it into large language models. She warns of the risk of synopsis leaks and relies on the AI committee of the Association of American Literary Agents, preparing guidelines for her authors, considering transparency essential. Keeping humans where value is created remains a priority. On translation and audio, the line is clear. Nora Mercurio from the German literary house Suhrkamp does not allow AI translations or AI narrators, while Laura Perciasepe from Summit Books, a brand of Simon & Schuster, assures that she has "no projects for AI-produced audiobooks." Adam Freudenheim from the independent British house Pushkin Press does not acquire "an audio created with AI, due to the inability to protect it with copyright," and never buys "a book without audio rights in the package." Regina Brooks nuances: AI is already part of the work of translators, but it is not yet able to handle interpretation, rhythm, humor, cultural contexts, or ambiguities. Proving authenticity in youth is another concern. Leonel Teti from the Hispano-South American group Urano World states that his audience is strongly opposed to AI: he had to repeat that none had been used, because teenagers and millennials "sometimes assume that even a digital drawing is generated by AI." His response is to publish videos of the illustrator creating the cover. Vicki Willden-Lebrecht from the Bright Agency advises her artists to "show off their brushes," and the British Liz White from the independent house Andersen Press notes the demand for "pictorial" images, meaning those with visible brushstrokes. In Vietnam, Quynh Huong from Slowbooks states that 90 percent of the illustrations from her house are hand-drawn. However, the country receives significant funding to test AI, and AI education has become mandatory even in schools. She fears that readers might copy the covers of her books into AI tools to create content that the house does not control, even though she notes that "legal protection is progressing very quickly." Uncertain legislation continues to be a challenge. The International Publishers Association (IPA) recalls that training AI on protected works falls under exclusive rights: companies must obtain licenses, and transparency conditions attribution and compensation. Out of 139 procedures initiated, the decisions made do not recognize fair use as a solid defense, and jurisprudence can be contradictory. While in France the National Syndicate of Publishers (SNE) has launched a legal battle against Meta and the "harvesting" of works to train its AI, in Germany, the Munich court ruled in November 2025, in a dispute between GEMA (the German equivalent of SACEM) and OpenAI, that the memorization of works by a model constitutes reproduction and that the exception of text mining does not apply. On the other side of the English Channel, the High Court of London ruled differently between Getty Images and Stability AI. The Court of Justice of the European Union was first approached and held its first hearing on March 10, 2026. It must determine whether these exceptions justify unauthorized training.