The Authors’ Licensing and Collecting Society (ALCS) has described the alleged use of “pirated software” by artificial intelligence (AI) company Anthropic to train its AI chatbot Claude as “outrageous” and “typical of a broader trend”.
Earlier this week, lawyers for authors Andrea Bartz, Charles Graeber and Kirk Wallace Johnson filed a copyright infringement lawsuit in California, claiming that Anthropic used “pirated” copies of their books to teach its AI chatbot Claude.
The lawsuit accuses Anthropic of “downloading and copying hundreds of thousands of copyrighted books from pirated and illegal websites.”
This is the latest in a series of legal cases accusing AI companies of using authors’ material to teach large language models (LLMs) without their consent or using allegedly stolen copies of their books.
The lawsuit against Anthropic accuses the technology company of committing a “brazen violation” of copyright by downloading “known pirated copies” of authors’ works and “feeding” them into its models.
The court documents say Anthropic took “these drastic steps” to help its computer algorithms “generate human-like text responses.”
The three US-based authors are demanding consent and compensation as well as damages “for the mass infringement of copyrighted works” and “an injunction to prevent such inappropriate behavior in the future.”
According to Anthropic’s latest financial statements, the company will generate revenue of over $850 million in 2024. The company has raised $7.6 billion in funding from backers such as tech giants Amazon ($4 billion) and Google ($2 billion).
Anthropic is a spin-off of OpenAI, the developer of ChatGPT. The latest version of Claude launched just 16 weeks ago, but reported gross mobile app revenue for iOS and Android of $1 million yesterday, according to a trading update.
A spokesperson for Anthropic told The Bookseller: “We are aware of the claim and are reviewing the complaint. We cannot comment further on ongoing litigation.”
ALCS Executive Director Barbara Hayes said: “Unfortunately, this is typical of a general trend where AI companies develop their models by using works of authors without permission, acknowledgement or compensation. At the same time, the revenue of the authors whose works are infringed is reduced. In the case of Anthropic, this seems even more egregious because the training data allegedly contains pirated content.”
ALCS recently conducted a survey of its members and received “an overwhelming number of responses from concerned authors” to explore licensing models that would give authors control over “how their works are used by AI companies and allow them to be compensated for such use if they so choose.”
In July, academics attacked Taylor & Francis (T&F) for selling access to its authors’ research as part of a $10 million partnership with Microsoft. Later, parent company Informa’s half-year results revealed that it would earn £58 million ($75 million) by selling access to its authors’ work to AI companies. Two other academic publishers – Wiley and Oxford University Press – subsequently confirmed that they have struck deals with, or are considering working with, artificial intelligence (AI) companies. Wiley said in its latest trading update that it had earned $23 million by giving an unnamed company access to its content to train its LLMs.
Ambre Morvan, Senior Policy and Public Affairs Manager at the Society of Authors (SoA), said: The bookseller: “The Society of Authors is very concerned about the threat that generative artificial intelligence (AI) poses to authors’ livelihoods and the future of the profession, and the potentially harmful effects of people reading and relying on AI-interpreted output. There have now been numerous reported cases in the UK and abroad of these systems being ‘trained’ and developed on datasets containing significant numbers of copyrighted works. We are monitoring these cases closely as we believe they demonstrate the urgency of the situation and the need for immediate action.”
