The AI Drake track, which mysteriously became viral over the weekend, is just the beginning of a problem for Google that could have a major impact in some way. It’s not yet clear how it will play out.
The basic facts are that there is a new song called “Heart on My Sleeve”, created by , a TikTok-user named @ghostwriter877. The vocals sound just like Drake and The Weeknd. The song appeared out of the blue over the weekend. has a few reasons to be suspicious.
The full version of the song was released after it went viral on TikTok. It is also available on Apple Music, Spotify and YouTube. Universal Music Group, the label of Drake and The Weeknd, issued a statement expressing concern about AI. The statement stated that using generative AI violates their copyrights. This is the statement from UMG’s senior vice president for communications James Murtagh Hopkins:
It’s a little mysterious what happened after. It is unclear what happened em>next/em>. The track was removed from streaming services like Apple Music and Spotify, who have tight control over their libraries and are able to remove tracks at any time. However, it remained on YouTube and TikTok which are platforms for user-generated content with established DMCA processes. One source with knowledge of the situation told me that UMG did not actually issue takedowns for the music streaming services, and so far hasn’t made any statements to industry trade publications. Drake and The Weeknd have not either. It’s strange – it seems as if Ghostwriter977 removed the track to create hype.
But , then TikTok as well YouTube removed the track. YouTube in particular pulled the track with a claim that it had been removed as a result of a copyright notification from UMG. This makes it a little more difficult and interesting for Google. To issue a takedown notice to YouTube, one must have… copyrights on something. UMG does not own “Heart on my Sleeve”, as it is an original. It’s also not a song from the label’s catalogue.
What did UMG say? The label has told me that is considered an unauthorized sample. It also claims that Metro Boomin’s producer tag, at the beginning of the song was a sample. A DMCA takedown notification was only issued for that specific sample. YouTube isn’t interested in pursuing the issue further. It’s not clear whether that tag was generated by AI or if it is a sample.
YouTube spokesperson Jack Malon explains the situation: “We removed this video after we received a valid notification of copyright for a sample that was included in the video.” “Whether the video was created using artificial intelligence doesn’t impact our legal obligation to provide a way for rightsholders who claim to have infringed their copyrighted content to remove it.”
UMG then issued individual URL-by URL takedowns on YouTube, as copies of this song appeared. I was told by a music industry source that UMG couldn’t use YouTube’s automated ContentID because it didn’t own it and couldn’t claim it to allow the system to match it. Ghostwriter977 , oddly, reuploaded on their YouTube page following the first takedown. It’s… still up. There’s still a lot going on.
If Ghostwriter977 uploads “Heart on my Sleeve” without the Metro Boomin tag they will start a copyright battle that pits Google’s future against that of YouTube.
All right? Now, here’s the issue: If Ghostwriter977 uploads “Heart on my Sleeve without the Metro Boomin tag they will start a copyright battle that could pit the futures of YouTube and Google in a zero-sum game. Google must either abandon all its AI-generated projects, such as Bard and the Future of Search, , or will piss major YouTube partners, like Universal Music and Drake. Let’s take a look.
A sample, or an interpolation to a melody is a “copy”. has been aggressively expanding music copyright, but the basis of it is still copies of songs. There’s no way to claim that Fake Drake is a direct copy of any Drake song. No copy.
UMG, Getty Images, and other publishers claim that copyright is violated when they collect all of the data needed to train the AI. They say that it is illegal to use Drake’s entire catalogue, every Getty image, every Wall Street Journal story, etc. to teach an AI how to create more Drake songs, photos, and news articles. The fake Drake songs produced by the AI would then be considered “unauthorized derivative works.” Phew! We’re still in the copyright realm that everyone knows. is pretending understands.
Google, Microsoft and StabilityAI are all claiming these training copies to be fair usage. By “fair”, they don’t mean “fairly determined based on an argument in a comment section of an internet forum,” but rather “fairly determined as determined by a judge based on a case by case application of 17 United States Code SS107, which lays down a four factor test for fair uses that is just as controversial and unpredictable as any other aspect in
When I asked Microsoft CEO Satya Nadel this question about the new ChatGPT powered Bing, he was not shy. He said, “At the end of it all, search is really about fair use.” In other places, the question of what constitutes fair use will have to be thoroughly thought out. There will be other times, when I believe there will be legal cases which must also set precedents. “
That’s because there is no actual precedent for saying that scraping data to train an AI is fair use; all of these companies are relying on ancient internet law Cases that allowed social media and search engines to exist. It’s messy and feels like those decisions will be up for grabs over the next decade.
Imagine that Google is a company that operates YouTube and is also racing to create generative AI like Bard. Bard is… trained using tons of internet data under a permissive fair use interpretation that will be challenged in a wave lawsuits. Universal Music Group releases a statement that is strongly worded about generative AI, as well as how it expects its streaming partners to respect the copyrights of its artists and its labels. What are you going to do?
- If Google agrees Universal’s claim that AI-generated songs are an unpermissible derivative work, based on the unauthorized use of training data and that YouTube should remove songs that labels flag as sounding like their artist, then it undermines its own argument about fair use for Bard, and for every other generative AI products it makes. It undermines the future for the company.
- Google may disagree with Universal and say that AI-generated music is acceptable because it’s just training an AI using existing works. This protects Google and its future, but could trigger a number of lawsuits in the future from Universal, as well as other labels. It also puts YouTube and Universal at risk.
Google’s Malon responded to my question about the dilemma by saying that “it is not YouTube’s responsibility to determine who “owns” rights to content.” We give both copyright holders and uploaders the tools they need to dispute any claims that are incorrect. If a dispute cannot be resolved by our dispute resolution process, it may need to be settled in court.
