Evidence of what an AI model retained, prepared for expert review and cross-examination.
IPLYR tests whether a specific AI model reproduces a specific work and documents the result so that counsel's retained expert can reproduce it, rely on it, and defend it.
- → A result for each work and model tested
- → Prompts and model responses preserved so the test can be re-run
- → Files sealed with a cryptographic hash
- → Limitations and error-rate status written into every report
- → The full method available to opposing experts under protective order
Where it fits in a matter
Before filing, testing can help decide which works and models to focus on. After an expert is retained, the report and its record become inputs to the expert's own analysis.
The question it answers is narrow on purpose: did this model reproduce this work under documented conditions. A narrow question is easier to defend.
What it does not do
A report does not decide infringement, fair use or damages, and IPLYR does not provide legal advice. A negative result is not proof that a work was never used in training. For closed commercial models, conclusions about training generally need material from discovery.
Why it holds up
The work builds on published research, every result can be reproduced from its record, and the limits are disclosed up front. An opposing expert will look for the weak points regardless; stating them first is what keeps the rest credible.
More detail: methodology and limits · frequently asked questions