Meta’s AI training on copyrighted content is ‘fair use’, US judge says

But creators could have grounds for complaint in future cases they can prove market dilution and unfair competition, judge adds

A US judge has said that Meta’s use of copyrighted books to train its AI models constitutes "fair use" under US copyright law. It follows a similar judgement about Anthropic earlier this week, and will come as disappointment to authors and other creators looking for compensation in what they see as use of their work without permission.

District Judge Vince Chhabria on Wednesday dismissed the case against Meta brought in 2023 by a group of authors including Sarah Silverman and Ta-Nehisi Coates, saying they failed to provide sufficient evidence that the market for their works had been diluted.

The ruling aligns with a decision in favour of Anthropic, where Judge William Alsup ruled that Anthropic’s training of its Claude LLM on copyrighted works was also "fair use."

Both judges emphasised the "transformative" nature of AI training, arguing that it does not “replicate or supplant” the original works but instead creates something new.

However, while Meta’s case focused solely on the legality of training AI on copyrighted materials, Anthropic faces a separate trial for allegedly downloading the works it used to train Claude from pirated "shadow libraries." This suggests that while AI training may be deemed fair use, the way the content was obtained may still present legal problems for AI companies.

For creators, these rulings will be disappointing. They suggest that AI companies may not need to seek permission or pay royalties for using copyrighted works to train their models. However, despite ruling in favour of Meta in this particular case, Judge Chhabria acknowledged the potential harm to the creative sector from GenAI, noting that it could "flood the market" with content, reducing the incentive for human creativity – in which case market dilution could be a valid complaint. He suggested creators reframe their legal arguments to focus on market dilution and unfair competition rather than direct copyright infringement.

For AI companies, the judgements will provide a significant boost, affirming their assertions that their training practices are legal, at least in the US, under fair use principles. But the outcome of the Anthropic case, in which the company is accused of using pirated materials, may give them pause, and Judge Chhabria’s comments about market dilution suggest that future lawsuits could succeed if plaintiffs can prove damage to their livelihoods.

In the UK, the government is reconsidering its stance on AI and copyright. Initially, it argued that requiring AI companies to seek permission from copyright holders would stifle innovation, but complaints from high-profile artists and members of the public appear to have led to a rethink.