News · 2026-10-28

New York court guidance keeps responsibility for AI-assisted filings with lawyers

New York’s court system has issued guidance telling attorneys to verify AI-generated legal authorities while retaining responsibility for the court papers they sign. The October 26 guidance permits AI assistance under existing duties and says there is no separate statewide AI disclosure requirement, while warning lawyers to check local rules and individual judges’ orders.

Key facts

• The New York State Unified Court System released the guidance on October 26.
• Its attorney guide is eight pages and addresses preparing court papers.
• The statewide approach permits AI use if existing obligations are satisfied.
• Primary sources are the official release and full guide.

The court system’s title identifies the intended audience and problem: “The Risks of Using Artificial Intelligence in Preparing Court Papers: What Every Attorney Should Know.” The issue is the reliability of a filed representation, not whether a drafting tool can produce fluent legal language. A plausible case name, quotation, or proposition still needs to correspond to an actual authority supporting the argument.

The associated model rule tells users to understand their tool, carefully review the filing, and independently ensure it contains no fabricated or fictitious material. Signing implicitly certifies that review. This preserves a familiar responsibility through a changed production process: a tool may help draft the document, but it does not become the person accountable for submitting it.

A concrete analogy is an assistant handing a lawyer a stack of index cards. Some cards might summarize real cases accurately; others could confuse two holdings or invent an authority altogether. A polished stack does not establish that the cards are reliable. The lawyer still needs to open the source, check the proposition, and decide whether it belongs in the filing. AI changes how quickly the stack arrives, not the standard that makes it usable.

The guide calls for checking generated cases, statutes, regulations, and quotations against authoritative sources. That requirement goes beyond determining that a cited page exists. A real decision can be quoted incorrectly or cited for an argument it does not support. Ground Truth’s natural-language-inference lesson explains the general distinction between a related source and a source that actually supports a claim.

The technical reason for the caution is straightforward. A language model learns patterns that help it generate likely continuations; those patterns do not automatically enforce a database constraint requiring every case name to exist. Ground Truth’s hallucination lesson explains why a highly plausible answer can remain false. Asking the same system whether it is certain is not equivalent to checking the original legal record.

The guidance also separates statewide policy from local practice. It says there is no separate statewide disclosure or certification requirement for AI use. That does not establish that every court accepts identical procedures. Lawyers are directed to examine local rules and individual standing orders, so a general statewide description cannot replace checking the requirements governing a particular filing.

The distinction matters for software procurement as well as drafting. A legal product may advertise citations or connected sources, but the court’s responsibility framework still requires the human filer to assess the resulting paper. Source links can make checking easier; they do not establish that the underlying authority has been correctly understood. The relevant product question is whether the workflow supports independent review, not merely whether its output looks professionally formatted.

There is a genuine tension in this approach. It allows useful assistance without imposing a blanket ban, but it places continuing demands on reviewers at the moment automation makes more text easier to produce. A busy lawyer could confuse a convenient review interface with adequate verification. Ground Truth’s automation-bias lesson describes why recommendations become easier to accept when they arrive confidently and fit the task.

A separate example of source discipline appears in Wikipedia’s AI-cleanup project, which emphasizes checking support for claims rather than relying on stylistic guesses about authorship. The institutions have different rules and purposes, but both illustrate a useful distinction: identifying how text was produced and determining whether its claims are supported are different tasks. That comparison is an analytical connection, not a claim that the court borrowed Wikipedia’s policy.

The dossier contains no verified independent expert assessment or measured effect of the new attorney guidance. It also does not establish new judge-specific duties from this document; its subject is attorney filings. Those limits prevent the announcement from becoming a broader claim that every legal use of AI has been resolved.

The policy’s immediate consequence is clear. AI assistance can fit within New York’s statewide filing framework, but the person submitting the work remains responsible for its accuracy and applicable procedural requirements. The useful test is whether each material authority survives independent checking, regardless of how quickly the first draft appeared.

Primary source, verified: read the paper →

Ground Truth. (2026, October 28). New York court guidance keeps responsibility for AI-assisted filings with lawyers. Ground Truth. https://groundtruth.day/news/new-york-courts-ai-attorney-guide-source-verification.html

@misc{groundtruth:new-york-courts-ai-attorney-guide-source-verification,
  title  = {New York court guidance keeps responsibility for AI-assisted filings with lawyers},
  author = {{Ground Truth}},
  year   = {2026},
  month  = {oct},
  url    = {https://groundtruth.day/news/new-york-courts-ai-attorney-guide-source-verification.html}
}

Topics: ai-policy · law · hallucinations · verification · professional-responsibility