The Anthropic settlement and the risks of a “use first, pay later” approach to AI and copyright

UK publisher Bloomsbury’s share of the $1.5 billion Anthropic copyright settlement is eye-catching, but the bigger story is what the case says about the value of creative work in the age of AI.

More than 14,000 Bloomsbury titles are covered by the US court-approved settlement, with $3,000 allocated to each work. Bloomsbury is only one of many rights holders involved, but its share shows that original content used to train AI systems has real commercial value.

It also raises a wider question: should AI businesses be able to build valuable technology using copyright-protected material first, and deal with rights and payment later?

The US judgment in Bartz et al. v Anthropic is mixed. Anthropic relied successfully on a US fair use defence in relation to works used for training its Claude AI model. However, the court found that works obtained from pirate sources and retained in Anthropic’s central library had been acquired unlawfully.

The debate around AI and copyright often gets pulled into technical questions about training, legal exceptions and new rules. But the underlying issue is much simpler: innovation should not rely on using protected work first and leaving rights holders to chase payment afterwards.

The US fair use defence does not apply in the same way in the UK – businesses in England and Wales cannot assume that the reasoning applied by a US court would provide the same protection. The legal position will depend on the UK’s copyright framework and how works are used.

The copyright challenge facing smaller creators

This is unlikely to be the last dispute of its kind. Journalists, photographers, musicians, software developers and designers all own works that may be valuable to AI developers.

There is also a practical imbalance between larger and smaller rights holders. Large publishers may have the resources to identify affected works and pursue claims. Smaller businesses, independent creators and individual authors may not know that their work has been obtained or used and may lack the evidence, funding or time to take action.

A copyright regime cannot work effectively if enforcing a right depends on the rights holder discovering an invisible use and then having the resources to challenge a much larger technology business.

That issue is central to the wider policy debate in the UK. The Government’s consultation on AI and copyright has considered approaches that could allow greater use of copyright-protected works for AI training, including models based on rights holders reserving their rights or opting out.

There are arguments in favour. AI developers need lawful access to high-quality content if the UK is to remain competitive in developing and adopting AI technologies. A framework that is too restrictive could make innovation more difficult.

But an opt-out model risks placing too great a burden on rights holders. Smaller businesses and individual creators may have limited ability to monitor how their content is being used, meaning responsibility would fall on the party least able to identify what is happening.

Why access and ownership must go hand in hand

The answer shouldn’t be to put the brakes on AI development. Businesses need access to data and content, while creators and rights holders need confidence that their work cannot simply be taken without meaningful control or compensation. Access and ownership are different things, and a workable framework needs to recognise both.

One possible route is a centralised national licensing system. This could give AI developers a practical route to access content lawfully while giving rights holders greater visibility and an opportunity to be paid. It would not eliminate disputes, but it could reduce uncertainty.

AI has enormous potential to help businesses innovate, improve services and operate more efficiently. But responsible innovation should not rest on the assumption that protected work is free simply because it is available online.

The Anthropic settlement is significant for more than its headline value. It is a reminder that the creative material underpinning AI systems has owners, rights and commercial value.

Protecting creative work and supporting responsible AI innovation are not competing objectives. The challenge is to make lawful access easier for AI businesses, while ensuring rights holders are not left to fight for compensation only after the value of their work has already been extracted.