How generative AI’s free-rider loop is testing the value of creation —from music and reference works to UGC platforms and beyond
By Lili Kazemi | General Counsel & AI Policy Leader, Anant Corporation
This piece continues my Anant series on AI intellectual-property battles—following
- From RAGs to Output Riches and
- When AI Creates Value, Settlements Aren’t Arm’s-Length—They’re Arm’s-Twist.
- Can We Stop Comparing Everything to Napster?
If Bartz v. Anthropic priced the past, these new lawsuits define the future.
Together, Concord v. Anthropic, Britannica v. Perplexity, and Reddit v. Perplexity trace the full scale of AI liability—from premium creative works to crowdsourced speech—showing how far the law can slide down the Circular Slop Scale:
Art → Knowledge → Conversation → Slop. So what comes first, the AI slop egg or the original clucking chicken?
🧭 Why We’re Here — The Circular Slop Thesis
This piece continues my Anant series on the AI intellectual-property wars, following From RAGs to Output Riches and When AI Creates Value, Settlements Aren’t Arm’s-Length — They’re Arm’s-Twist.
Together, these articles form a single inquiry: who sustains creation when machines can recycle it?
The answer is no longer theoretical. Each new lawsuit exposes a different part of what I call the Circular Slop Problem — the chicken-or-egg loop where:
- Human art feeds AI.
- AI outputs siphon attention and revenue from the original.
- Synthetic content floods back into the web and becomes training data again.
We’ve reached the moment when the loop is feeding on itself. And the three cases now in motion — Concord v. Anthropic (music), Britannica v. Perplexity (reference), and Reddit v. Perplexity (UGC) — map the entire scale of value erosion from premium art to public speech.
🎵 1 | Music — Premium Art and the First Domino
Case: Concord / UMG / ABKCO v. Anthropic (N.D.Cal. 5:24-cv-03811)
This case, which is one of the better known music copyright lyric cases, originally Concord / UMG / ABKCO v. Anthropic (No. 3:23-cv-01092), was formally transferred from the U.S. District Court for the Middle District of Tennessee to the U.S. District Court for the Northern District of California. The transfer was granted by the Tennessee court after Anthropic moved to dismiss for lack of personal jurisdiction or, alternatively, to transfer venue, arguing that the company is headquartered in San Francisco and the vast majority of its operations, including the creation and training of the Claude AI model, took place in California, making it the proper forum for the dispute.
UMG, Concord, and ABKCO continue to pursue Anthropic for allegedly training its Claude models on thousands of copyrighted lyrics and then outputting them verbatim without license. Despite a January 2025 guardrails stipulation, the case remains active. On October 6 2025, Judge Eumi Lee denied Anthropic’s motion to dismiss, allowing the publishers’ contributory and vicarious infringement claims to move forward. The court will now weigh discovery and damages.
Why it matters: Lyrics are the clearest test of output liability — short, distinctive, and commercially valuable. The case revives the Grokster¹ standard for secondary liability: if a platform knows infringement is occurring, benefits from it, and fails to act, intent can be inferred.
The domino sequence:
- Oct 29-30 2025 — UMG + Udio Settlement. UMG settled its copyright dispute with Uncharted Labs (d/b/a Udio), announcing a 2026 launch of a licensed AI-music platform. Under the deal, Udio disables downloads, adds fingerprinting, and pays opt-in royalties to artists and songwriters.
- Oct 30 2025 — UMG + Stability AI Partnership. One day later, UMG unveiled a strategic alliance with Stability AI to co-develop licensed AI-music tools for professionals.
The court has further ordered Anthropic to produce a “statistically significant sample” of Claude prompt and output records to gather more evidence for the case.
Signal: the music industry is pivoting from “block and ban” to “license and build.” UMG is still litigating against Anthropic even as it licenses other AI companies — defending its catalog while designing the next generation of authorized AI music.
📚 2 | Reference — Knowledge as Brand and Asset
Case: Encyclopaedia Britannica & Merriam-Webster v. Perplexity (S.D.N.Y., No. 1:25-cv-07546)
The complaint targets Perplexity’s “answer engine” for near-verbatim renderings of reference text and alleges trademark confusion under the Lanham Act when AI outputs appear to bear Britannica or M-W branding. The case is currently focused on the Motion to Dismiss, with the briefing schedule nearing completion. The court’s ruling on this motion will determine which, if any, of the publishers’ claims will proceed to the discovery phase.
Why it matters: This is the definitive “substitution” case for the RAG generation. If courts agree that AI answers replace rather than refer, then product design itself becomes a legal control surface: snippet length, link placement, and source labeling become compliance factors, not cosmetics.
💬 3 | User-Generated Platforms — Public Content Meets Private Control
Case: Reddit, Inc. v. Perplexity AI, Inc. et al. (S.D.N.Y., No. 1:25-cv-08736)
Filed Oct 22 2025, Reddit alleges Perplexity and co-defendants (SerpApi, Oxylabs, AWMProxy) used Google SERP excerpts to circumvent Reddit’s access controls and rate-limits — violating the DMCA anti-circumvention provisions² and undermining its API licensing business. The complaint adds unfair-competition and civil-conspiracy counts. Perplexity responded on Nov 3 2025, denying DMCA liability and asserting that its “grounding” via search is lawful under fair use and public-data principles.
Why it matters: This is the first AI case to apply the DMCA’s anti-circumvention framework to scraping through intermediaries. It tests where the open web ends and “controlled access” begins — the bottom rung of the scale where the commons becomes a battleground.
⚖️ What Bartz v. Anthropic Left Open
As I wrote in When AI Creates Value, Settlements Aren’t Arm’s-Length — They’re Arm’s-Twist, the Bartz settlement “priced the past but did not define the future.” That settlement addressed training data but avoided the harder issue of output liability — exactly what Concord, Britannica, and Reddit now confront.
These cases fill that vacuum:
- Concord → outputs and secondary liability for art.
- Britannica → substitution and brand integrity for knowledge.
- Reddit → circumvention and contract control for UGC.
Together, they define how far down the chain legal responsibility for AI’s “free-rider” loop can slide.
Five Questions for Counsel as the Dominoes Fall
- How far will the DMCA extend into AI grounding? If Reddit wins, scraping via search may constitute circumvention.
- Can UX be evidence of infringement? Britannica’s theory suggests yes.
- When does knowledge + benefit = intent? The Concord court’s Grokster reading revives secondary liability.
- Are settlements becoming regulation by contract? UMG’s Udio and Stability AI deals signal a new licensing order.
- Can culture reclaim its own feedback loop? If not, the Circular Slop may become permanent.
The Map of Creation
From songs to search, the loop is tightening. The music sector is licensing, reference publishers are litigating, platforms are fortifying — and the law is finally drawing its own map of AI’s value chain. Whether this ends in regulation, licensing markets, or creative attrition depends on what happens next in these cases.
But one principle is clear: AI still runs on human inputs — and human value must be paid forward.
¹ Metro-Goldwyn-Mayer Studios Inc. v. Grokster, Ltd., 545 U.S. 913 (2005). The Court held that a technology provider is liable for inducing infringement when it knowingly encourages or profits from infringing uses, a standard now invoked in AI-platform litigation such as Concord v. Anthropic.
² 17 U.S.C. § 1201 (Digital Millennium Copyright Act, 1998). The DMCA’s anti-circumvention rule prohibits bypassing “technological measures” that control access to copyrighted material—such as authentication systems, rate-limiting, or paywalls—even if the underlying work isn’t directly copied. In Reddit v. Perplexity, Reddit alleges that scraping Google search-result excerpts of Reddit posts unlawfully circumvented its own access controls and thus violated this section, a novel application of § 1201 in the AI-scraping context.
