In the legal dispute Thomson Reuters v. Ross, a US court decides for the first time whether feeding an AI with copyrighted content is permitted under the Fair Use Doctrine.
The District Court of Delaware says no, albeit not categorically, but on a case-by-case basis. In the present case, Ross is felled by the fact that it wanted to place its own competing product on the market with the help of information from a well-known US case law database. The decision is therefore not suitable as a “top or bottom” template for future cases. However, it opens up exciting perspectives and provides a breath of fresh air in the debate on AI and copyright. The decision is available here.
Background
Thomson Reuters, operator of the legal research platform Westlaw, owns an extensive collection of legal documents, including so-called headnotes, i.e. editorially compiled summaries of important US court judgements.
Ross Intelligence, a provider of AI-supported legal research, initially asked for a licence to use this content – an offer that was rejected. Instead, Ross had LegalEase create bulk memos from the headnotes to train its AI search engine. In December 2020, Thomson Reuters therefore sued AI startup Ross Intelligence, claiming that Ross had unlawfully used Westlaw’s copyrighted “headnotes” to develop an AI-powered legal search engine.
The court ruled in favour of Thomas Reuters that Westlaw’s headnotes did not merely enumerate mere facts but constituted creative, copyrighted content and held that Ross Intelligence could not rely on fair use.
Key points of the decision
The decision was one of the first to deal with the principle of fair use in the context of AI. Fair use is a legal principle that permits certain unauthorised uses of protected material, provided they serve the purpose of public education and the stimulation of intellectual production. It has a similar function to the limitation provisions of copyright law in Germany, such as the pastiche limitation or the text-and-data-mining limitation.
Decisive factors in the assessment of fair use are the purpose of the use (especially whether commercial or non-profit), the degree of copyright protection of the adopted works, the scope of the use and, finally, the impact on the market.
With regard to these relevant factors, Ross argues that only transformative use was intended, that they did not use complete copies of the headnotes and that there was no direct commercial harm to Westlaw.
However, the court does not follow this argument: it first recognises that the headnotes have the degree of originality required for protection – in addition to summarising court rulings, they also contain legal analyses prepared by expert editors. Furthermore, it stated that although the creative creation existed, it was to be classified as average in comparison to highly creative works. In addition, the amount of copied content was relatively small, as none of the copied headnotes appeared in the final editions of Ross’ search engine. These aspects indicate that, with regard to the degree of utilisation and the type of transformation, there are certainly circumstances that are covered by fair use.
On the other hand, however, the commercial purpose of the use weighed significantly more heavily. Following the case of Andy Warhol v. Goldsmith, the court found that Ross used the headnotes to develop a product that was in direct competition with Westlaw – a circumstance that emphasised the commercial nature of the use. The court was particularly critical of the fact that the tool developed by Ross was intended as a “market substitute” and thus had a negative impact on the potential market for licensed headnotes. Despite a possible public benefit in accessing legal information, the court does not see sufficient compensation to justify the economic damage to Westlaw. It also emphasised that copyright law serves to reward creators for their achievements, which is undermined by the unlicensed use of the headnotes.
After weighing these factors, the court concluded that Ross could not rely on fair use, so that the use of the headnotes constituted unlicensed and unlawful exploitation.
Conclusion
This specific decision may not be groundbreaking for all further decisions. For one thing, it is a judgement by a district court that can still be appealed. Secondly, unlike the proceedings pending against OpenAI, for example, the case is not about generative AI.
The decision at least confirms – for the USA – the protection of editorially created legal content and makes it clear that fair use requires a complex consideration of various aspects, with use for competitive purposes being particularly important.
Nevertheless, Europeans are certainly keeping a close eye on US case law, not least because many important proceedings relating to large AI tools such as OpenAI are pending. German courts could well follow aspects of US case law, even if fair use does not exist in the European legal area. The doctrine is not transferable.