An important AI copyright case to pay attention to if you are building AI models trained on data you didn't create or data you didn’t seek express consent to use.
Deep dive - the trial for AI copyright case Thomson Reuters Enterprise Centre GMBH and West Publishing Corp. (Thomson Reuters) v Ross Intelligence Inc (Ross) set to begin on 23 August 2024 has been postponed to allow for summary judgment motions on copyrightability, validity, infringement and fair use.
Unusually, Bibas J rescinded an earlier oral ruling where he appeared to find that Ross committed acts of infringement.
This case may have significant implications for organisations that produce content that is incorporated (potentially in breach of copyright) into AI models, and AI companies that develop and train AI models. It will bring more clarity as to copyright law’s attitude towards protecting content from AI and whether fair use will be a strong defence to wield.
Ultimately, it may have ramifications for the economic and legal viability of current approaches of extracting information for training AI models. Notably, many AI companies such as OpenAI and Stability AI are currently embroiled in copyright suits.
Thomson Reuters allege that Ross induced a third-party company to use its Westlaw account to copy Westlaw content unlawfully to Ross. Ross allegedly used the content to train its own ‘AI-driven...legal search engine’. The alleged copied content includes Westlaw’s Headnotes, key number organisation system and case summaries.
Given Thomson Reuters does not ‘own any of the underlying judicial opinions’ in its database, an important aspect of this copyright suit will be whether the content constitutes original works of authorship such that there is a ‘minimal degree of creativity’ (Feist Publications, Inc v Rural Telephone Service Company, Inc). Ross would have infringed copyright if Ross copied the parts that are original and thus protectable. For example, Ross could attempt to argue they copied the judicial opinions itself, rather than Westlaw’s content/derivation of the judicial opinions.
If it is held Ross did infringe copyright, it may be open to them to advance the fair use defence. This involves the court considering whether there has been transformative use, weighted against other factors such as whether it was for commercial use, the nature of the copyrighted work, the amount of the copyrighted work used, and the effect of the use on the market or value of Westlaw’s content. An interesting aspect to consider is whether the court will consider the usage of Westlaw’s content to train Ross’ AI tool is transformative.
Dorothy Sam Indira Blycha James Myint Ty Haberland Schellie-Jayne Price Monique Francis