AI Evals for Legal Domains & AI Agents | December 3, 2025 Brain Trust

A focused 90-minute session launching a 2026 research and practice agenda on custom AI evaluations for legal teams and organizations deploying AI agents.

Hosted from Stanford University on December 3, 2025, and co-convened by Stanford CodeX, Stanford HAI Digital Economy Lab, and law.MIT.edu.

📌 KEY THEMES
Why generic benchmarks aren't enough—your risks live in the specifics
Evaluation as governance: defining "what good looks like" for AI in your context
Evaluating AI agents: traces, process-level assessment, and observability
Measuring loyalty and fiduciary duty for consumer-facing AI agents
Building toward 2026: realistic evals, standards, and collaboration

🎤 BRAINT RUST SPEAKERS
Dazza Greenwood - Civics.Com (Host & Convener Opening Remarks)
Tara Waters — Vals AI (Legal AI Benchmarks)
Roman Engeler — Atla (Agent Measurement & LLM-as-Judge)
Darius Emrani — Scorecard (Agent Eval Infrastructure)
Dan Leininger — Consumer Reports Innovation Lab (Loyal AI Agents)
Robert Mahari — Stanford CodeX (Computational Law & Verification)

📚 RESOURCES & NEXT STEPS
Event page & updates: https://computationallaw.org
"* Beyond AI Benchmarks" (Dazza Greenwood): https://www.dazzagreenwood.com/p/beyo...
Vals Legal AI Reports: https://www.vals.ai/industry-reports

🔔 Want to learn more or get involved in the 2026 work? Use the form at ComputationalLaw.org to stay updated or inquire about collaboration.

—
Hosted and Convened by Daniel "Dazza" Greenwood
law.MIT.edu · Stanford CodeX · Stanford HAI Digital Economy Lab
Blog: https://dazzagreenwood.com | Consulting: https://civics.com