Harlequin used AI to translate romance novels. A first in France

About ten books from Harlequin's French division were translated using an AI tool in late 2025. Franceinfo called it the first case of its kind in the country. The publisher is known for its massive scale and fast production cycle, which makes it a natural place to test automation.

79 percent see a "threat"

A survey cited by the weekly Alternatives Économiques found that 79 percent of translators consider tools like DeepL and Google Translate a "threat" to their profession. Another 26 percent say they are already seeing fewer orders. Translators are among the groups most exposed to generative AI. As the wave grows, they try to build "dams" to protect their work. One mechanism is easy to miss. If a publisher gives a translator two months for four hundred manuscript pages, the deadline itself signals an expectation of a fast, superficial job done with a machine. Time pressure works like an order for automation, even when nobody says so out loud.

The "BrIAn" tool and a faster pace

Fluent Planet, a French company, developed the tool BrIAn. It offers the translator one or several versions of a rendering in English and French. Its chief executive, Thierry Tavakelian, insists that the human always has the last word. The publisher set two conditions: literary quality must not be compromised, and the translator's role remains central. The effect is measurable. The daily pace of work rises from a thousand to a thousand five hundred words to two thousand. Translators fear losing the status of author of the translation and seeing their name disappear from the cover. In exchange, they were offered the option of working for an outside subcontractor, and for less money. Samuel Sfez, president of the Association of Literary Translators, calls it a situation in which collaborations lasting twenty or thirty years come to an end, with no right to unemployment benefit. He appeals to publishers to drop AI in the translation of literary works. Joachim Schnerf, editorial director for foreign literature at Grasset, says plainly that what is at stake is the very notion of human creativity.

A court dispute over training

A dispute continues over what translation models learned from. In March 2025, the Syndicat national de l'édition (SNE), the Société des gens de lettres (SGDL) and the Syndicat national des auteurs et compositeurs (SNAC) sued Meta. They accused it of "counterfeiting" and "economic parasitism" over the mass use of protected works without consent to train a model. The lawsuit fits into a wider European dispute. Publishers and authors' organisations demand transparency about training datasets and payment for the use of their work. Those are the same demands the European Parliament adopted as recommendations in March 2026.

For the literary market, this means a double risk. On one side, lower rates and fewer assignments for translators. On the other, pressure on quality, because French readers are increasingly aware of what machine-generated text looks like. The debate about automatic translation of literature has stopped being a discussion about technology. It has become a dispute over who controls working time and the final shape of the language in a book.