The publisher of Harry Potter is about to cash a multi-million pound check from an artificial intelligence firm, and the reason should give every UK business owner pause for thought.
Bloomsbury, home to JK Rowling and Sarah J Maas and Susanna Clarke, has won a $1.5bn (£1.12bn) copyright deal between AI startup Anthropic and thousands of authors whose protected work was used to train its Claude chatbots.
The publisher said it has 14,087 titles listed for accommodation, with a proposed compensation of about $3,000 a title. After deducting around 10 per cent for lawyer’s fees and other costs, Bloomsbury and the authors involved can expect to receive around $19m (£14m), with the proceeds split between them.
The London-based company expects cash in installments, which are likely to start in the second half of this financial year.
For small companies, the important information is not the size of the Bloomsbury windfall but the principle: content has value, and AI models are built on it without payment.
Copyright has become the defining battleground of the AI boom. Tools like Anthropic’s Claude are trained on large amounts of data pulled from the open web, including copyrighted novels, articles and images. US AI companies have argued that this is allowed under the “fair use” doctrine, which allows the use of copyrighted work without the owner’s permission in certain circumstances. Creators, from global publishers to one-man studios, are increasingly not compliant, and want AI companies to seek permission first or at least pay for what they use.
The case began when novelist Andrea Bartz and two other authors filed a lawsuit in 2024. About 91 percent of the combined 482,000 jobs have now been accounted for. US District Judge Araceli Martínez-Olguín said the settlement provided “reasonable relief” to affected authors and publishers, while the authors’ lead lawyer, Justin Nelson, called it “the largest copyright recovery known in history”.
Anthropic’s deputy general counsel, Aparna Sridhar, said: “We are pleased that more than 91% of the authors and publishers involved in this settlement want their share of the payment, and we look forward to concluding this matter.”
It’s the first major settlement of the many AI copyright claims still pending in US courts, and it’s a sign of the broader battle now raging between publishers and AI developers.
For British companies, accommodation is at a critical juncture. Ministers backed away from a broad issuance of patents in AI training, preferring to allow a licensed market to develop. More than 90 percent of respondents to a government report on copyright and artificial intelligence support forcing developers to disclose the sources of their training data.
That leaves a trading opportunity lurking within the legal headache. Bloomsbury has already shown the way, announcing an AI license agreement last year that allows it to sell academic work to train manufacturing systems, with additional subject areas now under consideration. Authors are given the opportunity to “opt in” and get paid royalties if they allow their work to be used.
The study of small businesses that produce writing, images, code, video or research does not ask us: intellectual property is a commodity, not a product. Whether the company ends up defending it in court or licensing it for a fee, knowing what it’s up to, and whether it’s already been liquidated, it’s fast becoming a commercial enterprise.
Meanwhile, Anthropic is keeping a close eye on UK talent as it foots the bill for how its models are built. For businesses whose work has trained those models, the message from the California court is finally one they want to hear: pay up.
