HarperCollins has announced that it will allow some of its nonfiction backlist titles to be used for training artificial intelligence models, provided the authors give their permission. The publisher said in a statement that it has reached an agreement with an unnamed AI technology company to use a limited selection of these works to “improve model quality and performance.”
Authors have the choice to opt in to the deal or decline. The move follows revelations by US children’s author Daniel Kibblesmith, who disclosed on Bluesky that he was offered $2,500 (£2,000) for the use of his 2017 book Santa’s Husband, a work of fiction, despite HarperCollins stating the agreement covers nonfiction titles. Kibblesmith shared screenshots of an email from his agency detailing the offer, which included a three-year licence and restrictions on verbatim use per AI response.
The email described the terms as non-negotiable and claimed that several hundred authors had already agreed. It also noted that HarperCollins was required to keep the AI company’s identity confidential, though the agency expressed confidence it was “a major and respected company.”
HarperCollins defended the arrangement, stating that it has a “long history of innovation” and that its role includes presenting authors with opportunities while protecting their works’ value. The publisher emphasised that the agreement has “clear guardrails around model output that respects authors’ rights.”
In April, HarperCollins partnered with AI audio firm ElevenLabs to produce audiobooks for its foreign language business, describing AI as a “complementary tool” alongside voice actor-led productions.
