(Originally published by The Stanford Report on December 15, 2025.)

The era of AI evangelism is giving way to evaluation. Stanford faculty see a coming year defined by rigor, transparency, and focus on actual utility over speculative promise.

Legal AI turns to ROI, rigor, and multi-document reasoning

Julian Nyarko, Professor of Law and Stanford HAI Associate Director

I predict that two themes could define the year in the domain of AI for the legal services sector. First, rigor and ROI. Firms and courts might stop asking “Can it write?” and instead start asking “How well, on what, and at what risk?” I expect more standardized, domain-specific evaluations to become table stakes by tying model performance to tangible legal outcomes such as accuracy, citation integrity, privilege exposure, and turnaround time. There could also be a stronger focus on efficiency gains inside real workflows (document management, billing, and knowledge systems) rather than in controlled, artificial scenarios. Second, AI will take on harder work. Beyond intake and first drafts, we already begin seeing a shift toward systems that tackle, for instance, multi-document reasoning: synthesizing facts, mapping arguments, and surfacing counter-authority with provenance. This shift demands new frameworks for measurement – such as LLM-as-judge and pairwise preference ranking – to evaluate complex legal tasks at scale. Emerging benchmarks like GDPval, built around these ideas, could steer development roadmaps toward higher-order tasks.

Read the full article