AI in legal proceedings: Benjamin Wells & Ceyda Ilgen explore emerging risks & the practitioner’s duty
Generative artificial intelligence (AI) tools, general-purpose models and litigation-specific products alike, have moved from novelty to routine use in drafting, research and disclosure review in the space of two or three years. The upside is real: faster first drafts, quicker triage of large document sets and research assistance that can genuinely save time. The downside, now well documented, is that these tools can produce output that is fluent, confident, and wrong. The legal profession has already supplied several cautionary tales of what happens when that output is filed without being checked.
The hallucination problem
The now-familiar risk is that a generative AI tool, asked to find supporting authority, will produce a citation that looks entirely plausible—a sensible case name, a real-looking neutral citation, a quotation in the right register—but which does not exist, or exists but says nothing like what is quoted. The tool is not lying in any meaningful sense; it is predicting text that resembles the pattern of
