AI fears around second hand book sales
17 August 2026
It was only last month (July 2026) that the company Anthropic was ordered by a US judge to pay $1.5 billion in compensation to authors and publishers whose books had been pirated online and used to train the LLM (large language model) Claude. Now, according to a report from BBC News, there are fears that AI companies have changed tactics to train their LLMs and are buying up secondhand books to scan and then destroy.
The BBC report says that in America training AI with second hand physical books is lawful under US copyright law. In the UK, however, it isn’t: the permission of the copyright holder is required for a book is used to train AI. British second hand booksellers have reported receiving huge orders to be sent to warehouses overseas with their eventual destination unclear.
If American companies are trying to get around copyright law by buying books in Britain and avoiding paying for copyright to train LLMs by doing so in the US then this is an underhand move that will spark anger in the publishing community.
I searched The Atlantic’s LibGen database and found that my first three novels were pirated and most likely used to train AI without my consent or receiving any compensation for it. My fourth novel was published after the database snapshot was taken by The Atlantic in March 2025. Who knows if that has also been stolen and scraped?
Companies such as Google, OpenAI Anthropic and Meta have very deep pockets and to exploit authors rather than being honest and paying for copyright is morally and ethically wrong. A Society of Authors (SoA) survey showed that the majority of authors in the UK earn less than the British minimum wage annually and have to have another job to pay the bills. The fact that AI is increasingly being used to package stories put together from stolen works, putting writers out of a job (see Take A Break Fiction Feast magazine stopping paying writers and instead using AI to generate short stories) is the thin end of the wedge.
Whilst the UK government in March 2026 thankfully dropped plans, following lobbying by creative groups, to require authors to ‘opt-out’ rather than ‘opt-in’ to the use of their work in AI training, there is no imminent changes to copyright law planned to prevent LLM copyright abuse. Groups such as the SoA and Creators’ Rights Alliance campaign robustly but are Davids next to the power of the US Goliath giants.
Authors deserve, much, much better and this latest BBC News report will add fuel to the fire for the urgent strengthening of creators’ rights.