In (yet another) busy week for AI news, there have been two significant court rulings - one in Japan and one in the US - which respectively impact on two ongoing debates: do publicity rights allow people to protect their voice in the context of AI-generated replicas and is AI training ‘fair use’ under American law?
In the Japanese case, the Tokyo District Court said a performer could stop the unapproved use of their voice by AI by exercising their publicity right. Meanwhile, the US Third Circuit Appeals Court ruled that an AI company called Ross Intelligence could not rely on the fair use defence in its legal battle with Thomson Reuters.
Both judgments are potentially good news for the music industry, although the impact of the fair use ruling may be limited because it did not relate to generative AI.
Publicity rights in Japan
Starting in Japan, actor Kenjiro Tsuda - best known for voicing the character Kento Nanami in the anime series ‘Jujutsu Kaisen’ - sued TikTok over 188 videos posted by an anonymous account on the social media platform which, he said, included AI-generated narration that cloned his voice.
Both actors and musicians are keen to ensure they can stop the unapproved use of their voices in AI-generated content, and it’s generally thought the legal principle of publicity rights - also known as personality rights in some countries - can help in that regard. Although quite how that works in the context of voice is still being figured out.
This legal battle is the first time the question of publicity rights and AI-generated voice clones has been considered in a Japanese court, with presiding judge Aya Takahashi concluding that “the unauthorised use of a performer’s voice can be considered an infringement of publicity rights”.
Though, the judge clarified, that protection applies when “the purpose of using the voice is exclusively to take advantage of its commercial appeal”.
Which means a performer would need to show intent of that kind on the part of the person using an AI-generated voice clone, though in this case the TikTok account where the videos were posted used an image that resembled Tsuda’s ‘Jujutsu Kaisen’ character, presumably showing such intent.
Takahashi actually declined to issue an injunction forcing TikTok to remove the offending videos because they have already been deleted. But the ruling is still good news for actors and musicians in Japan.
That said, while it’s helpful when courts rule that publicity rights apply to a performer’s voice, in some countries performer groups are calling for specific digital replica rights to be introduced into the law which provide clearly defined protection in relation to both voice and likeness in the context of AI.
And securing such digital replica rights is particularly important in countries like the UK, where the concept of publicity or personality rights does not currently exist.
Fair use in the USA
Back in the US, the latest AI fair use ruling came as part of a long running legal battle between now defunct AI business Ross Intelligence and news and data organisation Thomson Reuters.
Ross developed an AI-powered legal-research search engine and as part of its AI training used content published by Westlaw - a legal industry platform owned by Reuters - specifically Westlaw’s ‘headnotes’ that summarise key points in legal cases. Ross’s AI tool then competed with Westlaw’s legal services.
Ross - like most AI companies accused of infringing copyright when training their models - claimed that its use of Westlaw’s content was fair use under American copyright law, meaning it didn’t need to get permission from Reuters to make copies of the headnotes.
Most of the lawsuits accusing AI companies of infringing copyright - including those filed by the music industry - swing on whether or not AI training is fair use. Copyright owners, like the record companies and music publishers that have sued Suno, Udio and Anthropic, insist AI training is never fair use. The AI companies argue that it definitely is.
Reuters v Ross became a high profile legal battle because it was the first AI fair use case to get to a judgement, first in a district court and now in a federal appeals court. And judges in both courts have ruled that, in the case of Ross and its AI-powered search engine, the use of Westlaw’s headnotes was not fair use. Meaning it infringed Reuters’ copyrights.
Any ruling that rejects the fair use defence of an AI company is useful for the copyright industries including the music industry. Which is why trade bodies representing record labels and music publishers in the US submitted an amicus brief to the Third Circuit Appeals Court urging it to uphold the lower court ruling in Reuters v Ross.
They wrote that this dispute “represents the first opportunity for a federal appellate court to determine what is being widely litigated throughout the country - whether copying copyrighted works for purposes of training a commercial AI model is subject to the defence of fair use when that model is used in a manner that competes with the copyrighted material on which it is trained”.
The district court, they added, “correctly answered that question in the negative”. And “while the factual contexts of AI legal research tools and AI-generated music may differ, the lower court’s holding establishes key precedent that training an AI model on copyrighted works to compete with and substitute for those copyrighted works can never be fair use”.
Needless to say, the labels and publishers will be pleased with the Third Circuit’s ruling - though, the fact the AI tool at the centre of this case was not a generative AI tool, like those developed by Suno, Udio and Anthropic, is in fact crucial.
And in more recent judgements in other copyright cases that did centre on generative AI models, developed by Meta and Anthropic, district court judges accepted the AI companies’ fair use defences.
The judge who originally ruled in the Reuters v Ross case explicitly noted this distinction in his judgement, writing that, “because the AI landscape is changing rapidly, I note for readers that only non-generative AI is before me today”.
And in a footnote in the new appeals court ruling, the judges note the recent intervention by the US Department Of Justice in a generative AI copyright case involving OpenAI, in which the government department argues in favour of AI training constituting fair use.
The judges say that the arguments presented by the DoJ “do not apply here” because “Ross’s AI platform cannot generate original expression”.
All of which allows Suno, Udio and Anthropic to argue that the fair use ruling in Reuter’s v Ross is not relevant to their respective legal battles with the music industry. Though, of course, the labels and publishers are likely to argue otherwise.