New York courts mandate human verification for generative AI filings

New York court officials have issued practice guidelines requiring legal practitioners to verify the accuracy of all generative AI output before submission.

The directive targets the growing adoption of large language models across legal drafting, research, and case analysis. Under the new guidance, attorneys retain direct responsibility for the accuracy of all pleadings, briefs, and supporting exhibits submitted to state tribunals. Algorithmic error will not excuse misstatements of law or fact.

The rules directly address the vulnerability of commercial generative models to hallucinated citations, distorted statutory text, and fabricated judicial opinions. Practitioners must independently cross-reference any AI-generated case citations with authoritative primary sources before incorporating them into filed arguments. The mandate extends beyond citations to factual summaries, analytical frameworks, and quotations attributed to courts or statutory codes.

State court leadership anchored the policy in longstanding ethical duties, including the requirements for professional competence, client confidentiality, and candour toward the bench. While the court system explicitly refrained from banning generative AI tools, it warned that unvetted reliance on automated text generation undermines court integrity.

Data governance forms a second core pillar of the guidance. Attorneys must evaluate how third-party AI platforms handle sensitive case materials. The framework highlights the risk that inputting proprietary client information into external models could violate confidentiality rules or waive attorney-client privilege. Legal teams must understand the training protocols, data retention policies, and security controls of any generative platform deployed in active matters.

The guidance also outlines the consequences of non-compliance. Practitioners who introduce false citations or unsupported legal propositions risk traditional judicial sanctions, including cost shifting, the striking of pleadings, and formal referrals to disciplinary committees. By clarifying that established professional conduct rules apply directly to synthetic drafting assistants, the court system aims to curtail automated procedural errors before they reach judges.

As automated drafting software becomes integrated into standard litigation workflows, the responsibility falls squarely on the bar to audit machine outputs. How effectively individual practitioners adapt their standard review workflows to catch subtle algorithmic errors before submission remains an open question.