Litigation and investigations involve large volumes of email, chat and documents. AI has been used here longer than anywhere else in law.
From TAR to generative AI
- Technology-assisted review (TAR), also called predictive coding, has been used for more than a decade: lawyers review a sample, the system learns what is relevant and ranks the rest. Courts have accepted it where the process is reasonable and transparent.
- Generative AI review goes further: you describe what is relevant in plain language and the model classifies documents, explains why, and summarizes key documents and themes. In The General Counsel Report 2026, 37% of general counsel reported using or experimenting with generative AI for first-pass review.
What to get right
- Defensibility: document the process, the validation (sampling to measure recall and precision) and who made decisions. Be ready to explain it to the court and the other side.
- Privilege review: AI can flag likely privileged documents, but final privilege calls stay with lawyers; use clawback agreements and, in U.S. federal courts, Federal Rule of Evidence 502(d) orders.
- Agreement with the other side: discuss the use of AI review in the discovery protocol early.
- Data handling: review platforms hold your most sensitive data; check where it is processed and whether any of it trains models.
Investigations
Internal investigations benefit from fast summaries and timelines, but they raise privilege questions: run them at the direction of counsel, inside approved tools, with access limited to the investigation team.