AI Governance & Quality Control
Attorney Supervision of AI Drafting: Practical Quality-Control Standards
By Anthony Villanueva ·
ABA Model Rules 5.1 and 5.3 establish the attorney's obligation to supervise AI-generated drafts. This guide translates those obligations into practical quality-control standards that can be implemented in a legal engineering workflow.
ABA Model Rule 5.3 establishes that a supervising attorney is responsible for the work of nonlawyer assistants, including technology systems that perform lawyer functions. The ABA's 2024 AI guidance makes explicit what this means for generative AI: attorneys who use AI drafting tools retain full professional responsibility for the accuracy, completeness, and appropriateness of the final work product, regardless of how that work product was generated.
"The AI wrote it" is not a defense to a professional responsibility complaint about an inaccurate or misleading legal document.
Minimum standards for drafting review protocols
Implementing effective attorney supervision of AI drafting requires converting a general professional obligation into specific, enforceable quality-control practices.
A minimum viable AI drafting review protocol includes: full read of every AI-generated document before transmission (not a skim, not a spot check — a complete read by a lawyer qualified in the relevant practice area); citation verification for every legal authority cited by the AI (the disciplinary cases involving AI-generated briefs with fabricated citations demonstrate what happens when this step is skipped); consistency check against the client's specific instructions and the relevant legal standards; and documentation of the review
(date, reviewer, issues identified, and changes made).
Importance of documentation in supervision
The documentation component is frequently omitted, but it is important for both quality control and professional responsibility purposes. A firm that can demonstrate that every AI-generated client deliverable was reviewed by a qualified attorney, with identified issues corrected before delivery, is in a fundamentally different professional responsibility position than a firm that relies on informal review without documentation. If a question arises later about the quality of the work product, the documentation trail is the defense.
Operationalizing quality gates in workflows
For high-volume applications — contract review at scale, due diligence in a large M&A transaction — the supervision standards need to be operationalized through a quality gate in the engineering workflow. The engineering system should produce an output that makes attorney review efficient and documented, not a raw AI output that requires the attorney to reconstruct context before they can effectively review it. Review interface design is a legal engineering responsibility, not an afterthought.
Frequently asked questions
What is the professional responsibility implication of using AI-drafting tools under Model Rule 5.3?
ABA Model Rule 5.3 establishes that a supervising attorney is responsible for the work of nonlawyer assistants, including technology systems that perform lawyer functions. The ABA's 2024 AI guidance makes explicit what this means for generative AI: attorneys who use AI drafting tools retain full professional responsibility for the accuracy, completeness, and appropriateness of the final work product, regardless of how that work product was generated.
"The AI wrote it" is not a defense to a professional responsibility complaint about an inaccurate or misleading legal document.
What specific steps are included in a minimum viable AI drafting review protocol?
Implementing effective attorney supervision of AI drafting requires converting a general professional obligation into specific, enforceable quality-control practices.
A minimum viable AI drafting review protocol includes: full read of every AI-generated document before transmission (not a skim, not a spot check — a complete read by a lawyer qualified in the relevant practice area); citation verification for every legal authority cited by the AI (the disciplinary cases involving AI-generated briefs with fabricated citations demonstrate what happens when this step is skipped); consistency check against the client's specific instructions and the relevant legal standards; and documentation of the review
(date, reviewer, issues identified, and changes made).
Who holds the responsibility for designing a review interface for AI drafting workflows?
For high-volume applications — contract review at scale, due diligence in a large M&A transaction — the supervision standards need to be operationalized through a quality gate in the engineering workflow. The engineering system should produce an output that makes attorney review efficient and documented, not a raw AI output that requires the attorney to reconstruct context before they can effectively review it. Review interface design is a legal engineering responsibility, not an afterthought.