Generative AI and Copyright Infringement: A Legal-Technical Analysis of AI Music Generation Systems Under 17 U.S.C. Title 17
This paper provides a legal-technical analysis of AI music generation under U.S. copyright law, arguing that while unauthorized lyric copying clearly infringes federal rights, voice imitation primarily implicates state publicity rights due to gaps in federal sound recording protection, necessitating new regulatory frameworks.
Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer
The Big Picture: The "AI Music Kitchen"
Imagine a high-tech kitchen where you can order a custom song just by typing a recipe into a computer. You can tell the computer: "Write these exact lyrics, but sing them in the voice of a famous pop star."
This paper asks a big question: If you do this, are you breaking the law?
The author, Zuhaib Hussain Butt, argues that the answer depends on what part of the song you are copying. The law treats the words (lyrics) and the voice (the singer's sound) very differently.
1. The Lyrics: The "Recipe" (Strictly Protected)
The Analogy: Think of a song's lyrics as a secret family recipe for a cake. The person who wrote the recipe owns it.
- What the paper says: If you type someone else's copyrighted lyrics into the AI, the AI is essentially copying that recipe. Even if the AI writes a brand new melody (a new way to bake the cake), using the exact words without permission is a violation.
- The Legal Risk: This is like walking into a bakery, stealing the owner's secret recipe card, and selling your own version of the cake. The law (Title 17) protects the "recipe" (lyrics) very strongly.
- The Verdict: If you use protected lyrics, you are likely infringing on the songwriter's rights. It doesn't matter if the AI wrote the tune; the words alone are enough to get you in trouble.
2. The Voice: The "Singer's Style" (The Legal Gray Area)
The Analogy: Now imagine you want the cake to be decorated to look exactly like a famous celebrity's face. You ask the AI to "make it look like [Celebrity Name]."
- What the paper says: Federal copyright law is surprisingly weak here. The law protects the specific recording of a voice (like a specific CD or MP3 file), but it does not protect the general "sound" or "style" of a voice.
- The "Sound-Alike" Rule: The law says you are allowed to hire a singer who sounds just like a famous star, as long as you don't steal the star's actual MP3 file and play it. It's like hiring a look-alike actor to play a role; that's legal, even if they sound exactly like the original person.
- The Verdict: If you use AI to clone a famous singer's voice without stealing their actual recording, you usually do not break federal copyright law. The courts have ruled that "sounding like" someone is not the same as "stealing" their recording.
3. The Loophole: State Laws and "Publicity Rights"
The Analogy: Since the federal "recipe book" (Copyright Law) doesn't stop you from using a celebrity's face or voice, some states have their own "bouncers" at the door.
- What the paper says: Because federal law leaves a gap, famous singers have to rely on State Laws (like the "Right of Publicity"). These laws say, "You can't use my face or my voice to sell things without my permission."
- The New Rules: States like Tennessee have passed new laws (like the "ELVIS Act") specifically to stop AI from cloning voices.
- The Verdict: You might not be breaking federal copyright law, but you could still be sued by the singer under state laws for "misusing their identity." It's a messy patchwork: in some states, you're safe; in others, you're not.
4. The Technical "How-To" (The Machine Behind the Magic)
The paper breaks down how the AI actually does this, mapping the tech to the legal risks:
- The Prompt: You type the lyrics. If they are stolen, the machine starts with a crime.
- The Brain (Transformer): The AI mixes the lyrics with a new melody. If the melody is too close to an original, it's a problem.
- The Voice Box (Vocoder): This is the part that makes the AI sound like a human. It uses "voice embeddings" (digital fingerprints of a voice). The paper notes that because the AI creates a new sound wave rather than copying an old file, the law often lets it slide.
5. The "Regulatory Gap" (The Missing Piece)
The author points out a weird imbalance in the law:
- Lyrics: Heavily protected.
- Melodies: Heavily protected.
- Voice Identity: Weakly protected (mostly by state laws, not federal).
The Problem: This creates a "perverse incentive." A user might think, "I can't copy the lyrics, but I can clone the singer's voice and make money off it." The paper argues this is unfair to the singers, whose unique voices are being exploited without a clear federal rule to stop it.
Summary of the Paper's Conclusions
- Don't steal the lyrics: If you type copyrighted lyrics into an AI, you are breaking the law (Copyright infringement).
- The voice is tricky: Cloning a voice isn't usually a federal copyright crime, but it might be a state law crime (Publicity violation).
- The system is broken: The law protects the words and the notes well, but it hasn't caught up with the "sound of a person."
- What needs to happen: The paper suggests we need new laws to protect voices federally (like a "Singing Voice Protection Act") and clearer rules for how AI companies can use data to train their models.
In short: The law is currently good at protecting the words of a song, but it's struggling to protect the singer when an AI mimics them. Until the laws change, the singer is the one most likely to lose out.
Drowning in papers in your field?
Get daily digests of the most novel papers matching your research keywords — with technical summaries, in your language.