AI testimony practice
AI Expert Witness for National and Federal Matters
Party-retained analysis and testimony on AI systems in federal court, MDL and arbitration: model behavior, training-data provenance, algorithmic bias claims, synthetic-media evidence and AI patent disputes.
The engagement
Testimony that stays inside what the AI record actually shows
This service is structured for counsel handling AI matters in federal district court, the Court of Federal Claims, the Federal Circuit, ITC 337 proceedings and multidistrict litigation. The technical work examines the model, the training and evaluation data available in the record, the inference outputs and the surrounding operational logs. Any opinion is bounded by the retained expert's qualifications, Federal Rule of Evidence 702 and the tribunal's orders.
AI cases live or die on what the record actually shows. A model card is not a training set. An accuracy claim is not a per-class error rate. A screenshot of a prompt is not a reproducible inference. The examination separates the artifacts produced by the AI system, the deployment and monitoring records that surround it, and the inferences that can and cannot be drawn from what is preserved. Every material finding is tied to a specific artifact and a specific tool version, and the report is written so a rebuttal examiner can locate the underlying evidence quickly.
The practice covers the AI subject matter categories active in federal courts: AI patent infringement and Markman construction where a claim reads on a training or inference process, trade secret claims under the Defend Trade Secrets Act involving model weights or training pipelines, algorithmic-bias disputes under Title VII, ADA, ECOA or FHA, FTC AI enforcement matters, generative-AI copyright disputes turning on training-data composition, and authentication questions for AI-generated or AI-modified evidence under FRE 901 and 902.
Scope
Model and training-data examination
Analysis of model weights, checkpoints, tokenizers, configuration files, training-data manifests, evaluation sets and versioned artifacts. Reproduction of the inference where the artifact record supports it, with tool versions, random seeds and hardware described in the report.
Algorithmic-bias analysis under federal statutes
Statistical review of model outputs against protected-class definitions used in Title VII, ADA, ECOA and FHA matters. Disparate-impact analysis, four-fifths rule application where relevant, calibration and equalized-odds review, and a written statement of the assumptions the analysis depends on.
AI patent and trade secret support
Markman claim-construction declarations, infringement and non-infringement opinions, invalidity and prior-art analysis under 35 U.S.C. sections 102 and 103, and trade-secret misappropriation opinions under the DTSA covering model weights, training pipelines, prompt libraries and evaluation harnesses.
Synthetic-media authentication for federal criminal and civil matters
Deepfake and AI-modified-media authentication under FRE 901 and 902, including provenance review (C2PA signatures where present), model-fingerprint analysis, generative-artifact analysis and chain-of-custody validation across preservation stages.
FTC and regulator-facing AI evidence
Technical evidence packaging for FTC AI enforcement matters, HHS OCR AI adjudications, DOJ civil rights division reviews and state attorney general actions. Documentation of the AI system, its deployment surface, monitoring evidence and the specific artifacts the enforcement question turns on.
Rebuttal, cross-examination and Daubert motion support
Independent review of an opposing AI expert's report, workpapers, notebooks and tool output. Rebuttal declarations, cross-examination question sets, Daubert motion declarations and demonstratives sourced to artifacts already in evidence.
Technical demonstratives for AI evidence
Timeline, artifact-trace and inference-reproduction demonstratives prepared as illustrative aids for counsel's use, subject to the tribunal's rulings. Each demonstrative is sourced to an artifact in the record and a stated method that a rebuttal examiner can run.
Methodology
How the AI expert engagement runs
-
Retention, conflict and scope
Retention letter with the retaining firm, a formal AI-matter conflict check across model families, dataset custodians and providers, and a written scope note identifying the AI questions the opinion will address and the artifacts the opinion will rest on.
-
Artifact reconciliation and reproduction
Reconciliation of the AI artifacts produced in discovery against the deployment and monitoring records that surround them. Where the matter turns on a specific inference, that inference is reproduced with the same tokenizer, weights and generation parameters, or the reproducibility gap is stated in the report.
-
Report drafting under FRE 702
A written expert report that ties every opinion to a specific artifact and a specific method. Bias findings state the statistical test, the protected-class definition, the sample and the assumptions. Patent findings state the claim construction the opinion assumes. Every reproducibility limitation is disclosed on the face of the report.
-
Deposition, hearing and trial testimony
Preparation with retaining counsel, deposition in person or by remote hookup, evidentiary hearing or trial testimony bounded by the report, and post-testimony support covering workpaper retention for appellate review.
Evidence commonly examined
Evidence reviewed
- Model weights, checkpoints, tokenizer files and configuration manifests
- Training-data manifests, evaluation sets and versioned dataset artifacts
- Inference logs, prompt histories, output-side records and monitoring exports
- Model cards, system cards, red-team reports and internal evaluation memoranda
- Deployment-environment records: container images, orchestration configuration, feature stores
- Opposing AI expert reports, workpapers, notebooks and tool output
- Court orders, protective orders and any ESI or AI protocol stipulations
What you can expect
What you receive
- Expert report or declaration under FRE 702 with per-finding artifact citations
- Reproducibility appendix stating tool versions, seeds, hardware and any deviations
- Bias-analysis appendix with protected-class definitions and statistical methods stated
- Deposition and trial testimony, in person or by permitted remote hookup
- Rebuttal declaration on an opposing AI expert's report and workpapers
- Daubert motion declaration and cross-examination question sets on request
- Post-testimony workpaper archive prepared for appellate review
Frequently asked
Common questions
Do you testify only for plaintiff or defense in AI matters?
The service may be retained by plaintiff or defense counsel, subject to conflicts, qualifications, scope and schedule. The opinion remains independent and stays inside what the record supports regardless of which side retained the expert.
Can you testify in ITC 337 proceedings and in federal district court?
Yes. The examiner has been retained in federal district court matters, multidistrict litigation, federal arbitration and ITC 337 proceedings that reach AI subject matter. Any appearance format follows the tribunal's rules and the terms of the retention.
How is model behavior reproduced when weights or seeds are not preserved?
Reproducibility limits are disclosed on the face of the report. Where weights, seeds or generation parameters are not preserved, the report says so, states what can and cannot be established from the surrounding record, and does not assert a reproduction that the artifacts do not support.
How does an AI opinion get through a Daubert motion?
Admissibility is decided by the court. What GDF supplies is a documented method: a stated theory or technique, the tests actually run, error considerations relevant to the finding, the peer-reviewed literature the method comes from and the examiner's own qualifications record. The method is designed so a rebuttal examiner can run it and get the same result.
Can you handle a matter that turns on training-data provenance where the dataset is disputed?
The examination can review dataset manifests, hashing evidence, retrieval logs, provider records in the produced record and preservation artifacts from the training pipeline. The report separates what those records establish from what would require additional discovery.
Do you produce rebuttal declarations against an opposing AI expert?
A rebuttal engagement reviews the opposing report, workpapers, notebooks and tool output. The rebuttal opinion is formed independently and may agree with, qualify or disagree with individual points according to the record.
What is a typical engagement structure for a national AI matter?
Conflicts, qualifications, scope, schedule, access requirements and deliverables are set in writing before work begins. Commercial terms are provided for the defined assignment and are not represented by the website.
Related capabilities
Related services
Talk with an examiner
Discuss the matter and the next step.
Call to discuss timing, scope and the safest way to share information. Do not send evidence or credentials by email.
24/7 hotline: 1-800-868-8189