Last week U.S. District Judge Vince Chhabria issued a set of 12 razor-sharp questions ahead of the summary-judgment hearing in Kadrey v. Meta—a lawsuit that claims Meta trained its Llama language model on “pirated” e-books. His order signals that courts are no longer asking whether AI training implicates copyright, but how deeply it cuts. Key themes include:
Why the Chhabria Questions Matter
- Pirated vs. lawfully acquired data – Does it matter if the training set came from shadow libraries? (Question 8)
- Market substitution – Could Llama’s outputs erode demand for the original books? (Questions 3–6)
- Transformative use – Is “training” itself transformative, or must we look at end-user outputs? (Questions 1–2, 10)
Reuters captured the judge’s tone succinctly: letting AI “obliterate” the market for original works is exactly what copyright law tries to prevent. 
Parallel Risks for Generative Music
Although the case involves books, the same fault lines run through music:
- Legal Pressure Point Why It’s Acute for Music
- Dataset provenance The best-sounding masters are almost always copyrighted.
- Market harm A synthetic track that’s “close enough” can cannibalize sync, streaming or performance royalties.
- Transformative threshold Courts haven’t decided whether “style cloning” of an artist’s catalog is transformative or substitutive.
How Aimi Designed Around the Minefield
Rather than argue after the fact that our training data is “fair use,” Aimi chose a build path that avoids the debate almost entirely:
- Artist-supplied, licensed source material only
We never ingested entire libraries of finished recordings. Instead, we model music production itself—notes, patterns, mix moves—so the system composes from first principles. - No wholesale scraping of commercial catalogues:
Every stem we do use comes from creators who have opted in under clear commercial terms. Revenues flow back through Aimi’s dashboard, so artists share in the upside instead of discovering infringement after the fact. - Granular provenance & audit trail:
Each musical gesture is traceable to a licensed source or to rules the model learned generatively, allowing us (and our customers) to prove ownership if a dispute arises. - Royalty-free outputs for users
Because inputs are clean, every soundtrack, mix or loop rendered by Aimi Sync or the Aimi Player API is automatically cleared for commercial use. That eliminates the takedown risk Judge Chhabria is worried about for text models.• Engage with artists and rights-holders early, before we write a line of code.
Looking Ahead
Whether the courts eventually bless broad “transformative” defenses or insist on blanket licensing, the music community will still need tools that respect creators while unlocking new forms of expression. By building an architecture that is clean by design, Aimi can focus on innovation—continuous, interactive, genre-perfect music—while sleeping well when judges start asking hard questions. Because in the end, transparency and trust beat litigation.
Extra Reading
You can learn more about how Aimi’s unique AI produces music by arranging, mixing, and mastering musical samples in real-time here: The AI Music Initiative.
Why is music AI particularly complex? Learn more here: Music AI @ Aimi.
Aimi feeds back into the artist economy, allowing artists to prosper while enabling creators, businesses, enterprises, app developers to take advantage of generative AI music solutions. Learn more here: Artists @ Aimi.