The internet was recently set abuzz with the emergence of a track titled “Heart on My Sleeve,” purportedly by Drake and The Weeknd. This song rapidly gained viral traction on platforms like TikTok and Spotify, captivating listeners with its catchy tune and lyrical style reminiscent of the two famous artists. However, the excitement was tinged with a significant revelation: neither Drake nor The Weeknd had any involvement in the song’s creation.
The track was, in fact, an AI-generated piece, the brainchild of TikTok user Ghostwriter977. This user had employed artificial intelligence, training it on the existing musical catalogs of Drake and The Weeknd to produce a completely new song that convincingly mirrored their distinctive voices, lyrical themes, and musical characteristics. The song’s virality was undeniable, amassing over 9 million views for the initial TikTok video. However, this success was short-lived. Universal Music Group, the record label representing both Drake and The Weeknd, swiftly issued copyright claims, leading to the removal of the song from TikTok, Spotify, and other platforms.
The swift takedown of “Heart on My Sleeve” has ignited a crucial debate about the legal and ethical landscape surrounding AI-generated art, particularly in music. Key questions have emerged: Did the AI-generated track infringe upon Drake and The Weeknd’s copyright? What legal recourse do artists have against AI-created music that mimics their style? And, does an AI-generated song possess its own copyright eligibility? To delve into these complex issues, Harvard Law Today consulted with intellectual property expert Louis Tompros, a lecturer on law at Harvard and partner at WilmerHale, offering insights into the nascent legal battles surrounding AI and copyright.
The Current Legal Landscape of AI-Generated Art
Harvard Law Today: What is the legal landscape right now related to AI-generated art?
Louis Tompros: Copyright law fundamentally grants copyright holders exclusive rights over their works. These rights encompass reproduction, distribution, performance, display, and the creation of derivative works. When we consider the impact of AI on copyright, it’s crucial to examine two primary areas. Firstly, the rights pertaining to AI-generated material itself. Secondly, the rights that individuals might assert against AI-generated content.
Addressing the first point, the central question is authorship and ownership of copyright in AI-generated material. Recent guidance from the Copyright Office offers significant clarity. On March 15th, they reaffirmed their stance: works autonomously created by AI, without substantial human intervention, cannot be copyrighted. This position is rooted in the U.S. Constitution, which empowers Congress to enact copyright laws, using the term “author.” Copyright statutes subsequently adopted “author,” consistently interpreted as a human creator. Therefore, copyright authorship, according to the Copyright Office, is exclusively human.
It’s important to acknowledge that the Copyright Office’s position, while significant, remains to be fully tested in courts. However, they have clarified a crucial nuance: works containing AI-generated elements can be copyrightable if they incorporate sufficient human authorship. For example, if a human creatively selects and arranges AI-generated components, the resulting work can be eligible for copyright protection.
HLT: And what about the second issue?
Tompros: The second major issue concerns the rights of human copyright holders when AI generates content that potentially infringes upon their work. This breaks down into “input” and “output” questions. Regarding “input,” the question is whether training AI models using copyrighted material constitutes copyright infringement. For instance, if an AI is trained by listening to vast amounts of music without the copyright holders’ consent, does this training process infringe on those copyrights? Or, is this use protected under fair use principles or other exceptions within copyright law?
Then, there’s the “output” question. Copyright law grants exclusive rights to create derivative works. If AI generates a work based on existing copyrighted material, is that output a derivative work that infringes upon the original copyright holder’s rights? Generally, creating music “in the style of” another artist is not considered a derivative work and is legally permissible. However, the use of machine learning and AI to generate such works introduces complexity. It remains unclear whether these AI-generated outputs are protected or infringe upon existing copyrights. Both input and output questions remain complex and legally unresolved.
The DMCA Takedown and Potential Legal Arguments
HLT: We saw the song get scrubbed from TikTok and some other places it initially appeared on the web, thanks to attorneys representing Universal Music Group. What are their best arguments for why it should be taken down?
Tompros: To understand the takedown, it’s essential to understand the Digital Millennium Copyright Act (DMCA) takedown process. The DMCA allows copyright holders to notify platforms like YouTube or Apple Music that content being distributed infringes their copyright. To maintain legal protections under the DMCA, these platforms are obligated to promptly remove the content. This is a DMCA takedown.
Reports suggest that Drake and The Weeknd’s attorneys utilized the DMCA to remove the “Drake Ai Song,” which makes procedural sense for a rapid takedown. However, the precise legal basis for the DMCA takedown request is not fully transparent. Some reports indicate the takedown wasn’t solely based on mimicking Drake and The Weeknd’s style, but rather on the alleged unauthorized inclusion of a “producer tag” by Metro Boomin within the AI-generated track. Producer tags are short audio signatures. If this tag was indeed copied into the AI song, it represents a direct audio copy, even if brief, potentially justifying a DMCA takedown based on copyright infringement of that specific audio element. However, the exact arguments presented in the DMCA notice remain unclear as these filings aren’t publicly available court documents.
Looking beyond the DMCA takedown, Drake and The Weeknd’s legal team could pursue broader copyright arguments, focusing on both the input and output aspects. They could argue that training the AI on Drake’s existing works necessitated unauthorized copying of copyrighted material to train the AI system, constituting copyright infringement. Additionally, they could argue that the AI-generated song itself is a derivative work, infringing on their exclusive right to create derivative works based on their original songs.
However, in my view, their strongest legal argument may not be copyright infringement, but rather a “right of publicity” claim. California and other states recognize a right of publicity, preventing unauthorized commercial exploitation of a celebrity’s identity. Landmark cases like Bette Midler v. Ford (1988) established that intentionally imitating a distinctive professional singer’s voice for commercial purposes violates their right of publicity under state law. In the Midler case, Ford used a sound-alike to impersonate Bette Midler for a Mercury Sable commercial after she refused permission to use her song. The court ruled this as a violation of her right of publicity.
This legal precedent appears directly relevant to the “Drake AI song,” where the AI intentionally imitated Drake and The Weeknd’s voices, potentially for commercial gain. The limitation of a right of publicity claim is the absence of a DMCA takedown mechanism. Drake and The Weeknd would need to initiate a state law claim, a more protracted legal process compared to a DMCA takedown.
Arguments Against Copyright Infringement
HLT: Based on the Copyright Office’s current guidance, the person who fed Drake and The Weeknd’s music into AI and asked it to create a new song from that material probably can’t be considered the song’s author. But are there arguments for why the song also didn’t violate those artists’ copyright rights?
Tompros: The creator of the “Drake AI song” could raise compelling arguments against copyright infringement. Firstly, regarding the input – the use of Drake and The Weeknd’s music to train the AI. While feeding copyrighted songs into the AI undoubtedly involves copying, the AI creator could argue this falls under fair use. Fair use, a provision in copyright law, balances copyright protection with freedom of expression under the First Amendment. It allows certain uses of copyrighted material without permission. The argument would be that AI training is transformative, not aimed at direct commercial copying. It’s about creating something new, and allowing AI to “learn” from existing music doesn’t directly harm the market for the original songs.
Secondly, concerning the output – the AI-generated song itself. If Drake and The Weeknd argue it’s a copy or derivative work, the AI creator can argue it’s a distinct creation. The lyrics are different, the music is different – it’s a new song, not a direct copy. If a human musician listens to Drake extensively and then writes an original song inspired by Drake’s style, but with different lyrics and music, it’s universally accepted as non-infringing. Why should an AI-generated song be judged differently?
AI Music vs. Music Sampling: A Comparison
HLT: In your view, how does AI-generated music compare to other industry-disrupting technologies like music sampling?
Tompros: Music sampling, when it emerged, also raised significant copyright questions, though arguably more straightforward than AI. Sampling unequivocally involves copying. Directly taking a musical riff and incorporating it as a sample is a clear act of copying. The legal question then becomes whether this sampling constitutes fair use. This often hinges on the amount sampled, its necessity, and how transformative the new work is compared to the original.
The Supreme Court case involving 2 Live Crew’s sampling of Roy Orbison’s “Pretty Woman” provides a relevant precedent. 2 Live Crew sampled a recognizable portion of “Pretty Woman,” and the court ultimately deemed it fair use because it was transformative and used only the necessary amount to reference and reinterpret the original.
AI music generation differs from sampling. AI training involves processing entire musical works, not just snippets. The AI “learns” from the entirety of Drake’s musical catalog. However, the output is entirely newly generated; it doesn’t directly incorporate sampled segments from the original works. This makes the fair use analysis more complex for AI music. Nonetheless, AI is undeniably disrupting the music industry, and the legal system needs to adapt and clarify these complex issues.
“The balance between artists’ rights in protecting what they have done and artists’ rights in creating something new is exactly the balance that has underlaid copyright law since it was embedded in the Constitution.”
Balancing Artist Rights and Innovation
HLT: How do we go about balancing the protection of artists’ rights to their own materials with promoting innovation in music?
Tompros: The tension between protecting existing artist rights and fostering new creative expression is inherent in copyright law since its constitutional foundation.
From a market perspective, the goal is to encourage new music creation. We must consider what best incentivizes musical innovation and where to draw the line in protecting existing music to encourage future creation. Economically, would Drake have invested in creating and distributing his music if he knew AI could replicate his style? It’s a complex question. Creatively, artists perceive their work as personal property and seek control over it. Simultaneously, free speech rights are vital. Copyright shouldn’t stifle commentary, innovation, or artistic evolution building upon existing works.
A balance is essential. For a century, copyright law has strived to adapt to technological advancements, maintaining incentives for creators while protecting artists’ rights.
Needed Legal Changes for AI-Generated Art
HLT: What, if any, legal changes do you think are needed to address AI-generated art?
Tompros: Ideally, a thoughtful statutory amendment to the Copyright Act specifically addressing AI-generated art would be beneficial. However, significant federal legal changes are challenging, particularly given the need for international harmonization of copyright law. Therefore, substantial AI-focused Copyright Act revisions are unlikely in the near term.
More realistically, legal evolution will likely occur through court decisions as they grapple with AI-related cases. Continued regulatory guidance from the Copyright Office is also expected. Key areas requiring judicial clarity include: Firstly, judicial agreement or disagreement with the Copyright Office’s stance on AI authorship. Secondly, judicial consensus on whether using copyrighted material for AI training constitutes fair use. Thirdly, clearer legal definitions of whether AI-generated works “in the style of” existing artists are considered derivative works.
Alt text: Louis Tompros, Harvard Law lecturer and intellectual property expert, discusses the legal complexities of AI-generated music.
Alt text: Cover art depicting Drake and The Weeknd, associated with the viral AI-generated song “Heart on My Sleeve”.
