Technology litigation

AI and machine learning litigation

Expert testimony about models, training data, and evaluation — written by a reporter who prepared the vocabulary before the deposition started.

Request a quote Technology litigation

Subject matter

What these depositions cover

The disputes differ; the transcript problem is the same. Every one of these produces long technical answers that have to be citable without a bracketed correction in the middle.

Training data and provenance

Where a dataset came from, how it was collected, licensed, filtered, and deduplicated, and who decided each of those — testimony dense with source names, dates, and version history.

Model architecture and training

Architecture, parameters, training procedure, fine-tuning, and the engineering decisions behind them, including the internal names a team used that appear nowhere in public documentation.

Evaluation and performance claims

Benchmarks, metrics, error analysis, and the gap between a marketing claim and a measured result — where a misheard metric name changes what the answer says.

Case types

Where AI testimony shows up

These matters arrive under several different legal theories and all of them end up in the same room with the same expert.

Intellectual property

Patent and trade secret disputes over model design, training pipelines, and inference systems, frequently under a source code protective order.

Copyright and data rights

Disputes over training corpora and generated output, where provenance testimony and licensing history carry the case.

Algorithmic bias and consumer claims

Employment, lending, and consumer-protection matters involving automated decision systems, combining statistical testimony with regulatory framing.

AI and machine learning

Questions about these cases

What makes an AI or machine learning deposition hard to report?
Density and novelty at the same time. A single answer can carry a model name, a library, a version number, an evaluation metric, and a parameter, none of which appear in ordinary usage. Those terms have to be in the dictionary before the deposition, because there is no time to resolve them mid-answer.
Do you report depositions about training data?
Yes. Data provenance testimony — where a dataset came from, how it was collected, licensed, filtered, and deduplicated — is common in both copyright and trade secret matters, and it is heavy with source names and dates that need to be exact.
Can you handle model architecture and evaluation testimony?
Yes. Architecture, training procedure, benchmark and evaluation methodology, and error analysis are ordinary expert subject matter in these cases. The preparation is a glossary built from the expert reports and the asserted materials.
What about algorithmic bias and disparate impact claims?
Those depositions combine statistical vocabulary with employment or consumer-protection framing, and the statistical terms are the ones most often garbled. They go into the dictionary alongside the technical ones.
Does any of this mean you use AI to produce the transcript?
No. The record is written stenographically by a certified reporter and certified by that reporter. Artificial intelligence is the subject matter of these cases, not a substitute for the reporter taking them.

Deposing a machine learning expert?

Send the reports and the witness list with the notice. The model names, libraries, and metrics are in the dictionary before the first question.

Request a quote