Attorneys must follow an eight-step protocol before using AI on client matters

In April 2026 a Sullivan & Cromwell attorney submitted an AI‑generated brief with fabricated citations, highlighting a broader trend of AI hallucinations that has already cost courts over $145,000 in sanctions and led to license revocations. The problem is not the technology itself but the lack of a structured workflow. Jennifer Case, J.D., founder of Law Tech AI, developed an eight‑step Legal AI Protocol that gives attorneys a defensible, repeatable process. The protocol begins with a foundation step that requires paid, enterprise‑grade AI accounts with data‑processing agreements, followed by careful tool selection—choosing purpose‑built legal AI for research and general‑purpose models for drafting. Prompting is guided by the CLAR framework (Context, Legal framework, Assignment, Requirements) and includes guardrails to prevent fabricated citations and to flag uncertainty. Research workflow mandates primary research in verified databases such as CoCounsel Legal, followed by a Retrieval‑Augmented Generation approach for synthesis, and a rigorous verification layer that requires reviewing every AI‑identified citation and running KeyCite and Quick Check before filing. Supervisory responsibility remains with the attorney, who must review the final document, disclose AI use when required, and ensure compliance with local rules. By following this eight‑step protocol, lawyers can harness AI’s speed and accuracy while protecting client confidentiality, maintaining professional standards, and avoiding costly sanctions.

(Source：Complete Ai Training)