Evidence for creative work in the age of AI.
IPLYR measures whether AI models reproduce specific copyrighted works, and documents the result so your retained expert can stand behind it.
Sources: Authors Guild and Susman Godfrey on the Anthropic settlement; ChatGPT Is Eating the World AI Copyright Case Tracker; European Commission, AI Act Article 53. IPLYR is not counsel of record in these matters.
Does that model reproduce this work?
IPLYR tests whether a specific AI model reproduces a specific copyrighted work, then documents what the test showed and what it cannot show. The record is built so your retained expert can re-run it, explain it and stand behind it.
When a model does not reproduce a work, the report says exactly that. A result like that is not evidence that the work was absent from training, and every report states this plainly.
Our approach builds on published research on extractable memorization. The full method is shared with retained experts, and with opposing experts under protective order. Whether any evidence is admitted is for the court to decide.
We work for rights holders and the people who represent them.
Litigators, the experts they retain, and the publishers, estates and authors whose works are at issue.
Per-work evidence for a live matter
We test the works at issue against the models at issue and deliver a documented record your retained expert can re-run, explain and testify to in their own words. We work through counsel.
- Scoped to your works and the models in suit
- What a result does and does not show, stated in every report
- Conflict-checked before we receive confidential material
Know what the models reproduce
We test whether AI models reproduce your works and report what that does and does not show. A result is most useful inside a legal or licensing strategy, so we work alongside your counsel.
- Per-work results with their limits stated
- Delivered to you and your counsel
A record your expert can stand behind.
Each engagement is scoped with counsel before any testing starts.
A written report
What was tested, how it was run, what it showed, and what it cannot show.
A record that can be re-run
Enough detail for your retained expert, and for an opposing expert under protective order, to repeat the test and check the result.
Tamper evidence
The report's files are sealed with a cryptographic hash, so a later change to any of them can be detected.
Technical details are shared with retained experts, and with opposing experts under protective order. Chain of custody for the matter remains with counsel.
The rulings shaping this field.
Short summaries of decisions that matter for claims about AI training and AI output.
Bartz v. Anthropic
Training on purchased books was fair use; a library of pirated copies was not. The case settled in 2025 for $1.5B.
Kadrey v. Meta
Fair use on summary judgment where plaintiffs failed to develop a market-harm record.
GEMA v. OpenAI
Memorization in the model was held to be a reproduction under German copyright law.
Holdings are summarized. Consult the full opinions. Nothing on this site is legal advice.
Bring us a work and a model. We will tell you what the evidence shows.
IPLYR is a forensic measurement firm. We do not provide legal advice or represent parties in litigation.