17.14 Legal
Law has been an early adopter of NLP for document review and is now an active deployment site for LLMs.
Contract review and analysis. Tools including Kira, Luminance, Lexion (acquired by DocuSign in 2024) and Harvey AI (founded 2022, raised $300M+ Series C in 2024) have moved law-firm contract review from junior-associate manual work to ML-assisted workflow. Allen & Overy was Harvey's first major law-firm partner; by 2025 most magic-circle and white-shoe firms had adopted at least one LLM-based assistant.
E-discovery. Predictive coding (technology-assisted review, TAR) for litigation document review predates LLMs by over a decade and is now court-accepted in major jurisdictions. The 2012 Da Silva Moore v. Publicis Groupe ruling in the Southern District of New York was the first US opinion explicitly approving TAR. LLMs have substantially improved the quality of document classification and have made narrative summarisation of large document collections feasible.
Legal research. Westlaw and LexisNexis have integrated generative AI; Thomson Reuters' CoCounsel (acquired from Casetext in 2023) provides an LLM legal-research assistant. Open-source efforts including Pile of Law and various law-fine-tuned LLMs are also active.
Hallucination risks in court. The 2023 case Mata v. Avianca (Southern District of New York) became the canonical cautionary tale: lawyer Steven Schwartz submitted a brief citing six entirely fabricated cases generated by ChatGPT, was sanctioned $5,000 and reported widely. Subsequent cases involving the same failure mode, fabricated citations, misquoted holdings, hallucinated precedent, have continued to surface, and most US federal courts now require lawyers to certify that AI-generated content has been verified. The lesson generalises: LLMs in high-stakes professional contexts must be retrieval-grounded, the retrieval must reach authoritative sources, and the human professional retains responsibility for verification.