Kadrey v. Meta
2023-Authors alleged LLaMA was trained on Books3. The fight turned on whether plaintiffs could demonstrate per-work memorization, beyond mere inclusion in a dataset.
Evidentiary gap → per-work memorization proof
The problem
Plaintiffs are losing winnable cases for lack of one thing: forensic evidence.
Discovery surfaces probabilities. Trials demand proof. Without reproducible methodology and a defensible chain of custody, even strong matters collapse at Daubert.
Expert engagements run $50K-$500K and 6 to 12 weeks. IPLYR delivers the same evidentiary tier for $2K-$5K in 7 days.
Getty v. Stability AI
2025At trial in the UK High Court (November 2025), Getty dropped its primary training claims. The court rejected the secondary infringement claim, holding that the model weights are not an infringing copy, and made limited trademark findings tied to watermarks generated by older model versions.
Evidentiary gap → evidence of the work inside the model
Bartz settlement context
2025Anthropic settled for $1.5B, about $3,000 per work across about 482,000 works. Per-work figures of this kind make reproducible, work-level attribution central to valuing future claims.
Evidentiary gap → reproducible methodology
Landscape
How IPLYR compares.
Independent landscape analysis of the AI training-data attribution space.
| Capability | IPLYR | ProRata | Patronus | Spawning | Story |
|---|---|---|---|---|---|
| Funding | Solo / pre-seed | $30M | $17M | $6M | $54M |
| Daubert-tier evidence packaging | |||||
| Memorization forensics | 20 documented methods | Hallucination focus | Opt-out registry | ||
| Jurisdictions modeled | 9 + 10 US states | US | US | US | US |
| Pricing | $2K-$5K / report | Enterprise | Enterprise | Free opt-out | Token-based |
Forensic evidence is a category none of them serve. Plaintiffs in Kadrey, Getty, and Bartz-style matters are losing for lack of exactly this.
IP
Founder, IPLYR
Open to expert-witness collaborations
"I built IPLYR because I sat through too many AI copyright matters where plaintiffs had a real case and no forensic vocabulary to prove it.
The methods exist in the literature. Min-K%++, nv-recall, Cooper PDE, TRAK, MIA ensembles, but they live in conference papers, not in litigation workflows. IPLYR translates them into evidentiary-grade packages with chain of custody, Daubert classification, and jurisdiction-aware analysis.
Open to expert-witness collaborations and academic partnerships."