As generative AI has made its way into the hands of consumers in recent months, the music industry has stepped up efforts to protect its intellectual property against copyright infringement – even if the line between fair use and infringement isn’t always clear in the new AI ecosystem.
One question that has come up is whether using a copyrighted work to train an AI model is itself a violation of copyright. For instance, if an AI model learns music by listening to the works of Ed Sheeran, and then incorporates Sheeran’s musical or vocal stylings into its own creations, is that copyright infringement?
In the view of many in the music industry, yes it is.
After an AI-generated track purporting to be from Drake and The Weeknd went viral on TikTok, YouTube and elsewhere in April, Universal Music Group (UMG) was unequivocal in its assertion that ripping off a musician’s sound and style amounts to infringement.
“The training of generative AI using our artists’ music (which represents both a breach of our agreements and a violation of copyright law) as well as the availability of infringing content created with generative AI on DSPs, begs the question as to which side of history all stakeholders in the music ecosystem want to be on: the side of artists, fans and human creative expression, or on the side of deep fakes, fraud and denying artists their due compensation,” UMG – which works with both Drake and The Weeknd – said in a statement to MBW.
UMG added that it was “encouraged by the engagement of our platform partners on these issues,” referring to the fact that major social media sites pulled the “fake Drake” track from their services after presumably receiving takedown requests.
Now, another test of copyright law in the age of AI is in the offing, this time focused on social platform Discord, and a server on its platform where users have been sharing guides and AI models used in the generation of music – including datasets of entire copyrighted songs.
The Recording Industry Association of America (RIAA) has sent a DMCA subpoena to Discord, asking the platform to provide it with the identities of certain individuals who have been sharing musical data on the “AI Hub” Discord server.
The subpoena, filed with the US District Court for the District of Columbia on June 14, requests that Discord “disclose the identities, including names, physical addresses, IP addresses, telephone numbers, e-mail addresses, payment information, account updates and account histories” of users who had posted Discord links featuring datasets of copyrighted songs.
Among the songs the RIAA says were shared without authorization are Mariah Carey’s Always Be My Baby, 50 Cent’s In Da Club, Olivia Rodrigo’s Jealousy, Jealousy, Shakira’s Whenever, Wherever, and Frank Sinatra’s In The Wee Small Hours of the Morning.
Discord has until June 30 to comply with the subpoena, or contest it in court.
“Unlike its predecessors, much of the latest generative AI is trained on copyrighted material, which clearly violates artists’ and labels’ rights.”
Sir Lucian Grainge, Universal Music Group
However, the RIAA is also going a step further than that.
The subpoena to Discord was preceded by two letters from the RIAA to Discord’s copyright enforcement team, one dated May 26 and the other June 12, which requested that Discord disable the server altogether.
“Specifically, we request that you (i) remove and/or disable access to this Discord server, (ii) remove the files or links from your system, including any mirrored or duplicate copies of those files or links, and/or that you disable all access to the infringing files and associated links, and (iii) inform the server operator/s and the uploader/s to that server(s), as applicable, of the illegality of their conduct,” the RIAA’s May 26 letter to Discord stated.
You can read the subpoena and the RIAA’s letters to Discord in their entirety here.
So far, Discord has not complied with the request to shut down the server, which currently has more than 142,000 members. According to a report at digital piracy news site Torrentfreak, subscribers to Discord’s AI Hub server use it not only to share datasets of copyrighted songs, but also to share AI guides and AI models, including voice models of well-known musicians.
Whether or not the RIAA is claiming that these voice models are also infringing works is unclear; their takedown request lists only datasets of entire copyrighted songs.
But the RIAA’s member recording companies may be feeling a sense of urgency about the server and its potential to become a major new source of AI-related piracy. According to Torrentfreak, the number of members on the AI Hub server has nearly doubled in just the past month.
And it’s also clear that many in the music industry consider it a violation of copyright law to train AI models on copyrighted music without authorization. UMG’s Chairman, Sir Lucian Grainge, made this clear in comments during the company’s Q1 earnings call earlier this year.
“Unlike its predecessors, much of the latest generative AI is trained on copyrighted material, which clearly violates artists’ and labels’ rights and will put platforms completely at odds with the partnerships with us and our artists and the ones that drive success,” Grainge said.
“Should platforms traffic in this kind of music, they would face the additional responsibility of addressing a huge volume of infringing AI-generated content.”
However, the laws surrounding copyright infringement by AI are still being worked out. The US Copyright Office has launched a public consultation on how copyright laws should be enforced with respect to AI, while US Congress members are currently holding hearings on regulation of AI.
Meanwhile, both the European Union and China have moved towards some restrictions on how AI uses copyrighted material.
The EU’s AI Act – which moved a big step closer to becoming law when the European Parliament voted in favor of it earlier this month – requires developers of “foundational” AI models (meaning AI models on which other AI models can be built) to declare whether copyrighted material was used in the training of the AI.
China, which has advanced two sets of AI regulations over the past half year, has proposed rules that would prohibit copyrighted materials from being used without permission.
However, one jurisdiction appears to be going in a different direction. Japan’s government, which has made it a goal to make the country a major player in the development of AI, has suggested that Japan’s copyright laws don’t forbid AI from learning off of copyrighted materials.
That’s true even if that material exists for expressly commercial purposes, so long as it isn’t reproduced, Japan’s Minister for Education, Culture, Sports, Science and Technology, Keiko Nagaoka, said at a public hearing in April.
However, it remains unclear in most jurisdictions today where the boundary between fair use and copyright infringement lies in the word of AI. The RIAA’s campaign against the Discord AI Hub server could prove to be an opportunity to test those boundaries.Music Business Worldwide
