New York Courts AI Guidance Lets Lawyers Skip Disclosure

Estimated reading time: 7 minutes · Last updated: 28 October 2026

New York's court system has told lawyers they may use generative AI to draft legal filings without telling the judge. The condition attached to the New York courts AI guidance is accountability. An attorney who signs a filing still certifies that it holds nothing false, and faces the same sanctions as before if a chatbot invented a case. The court system's artificial intelligence advisory committee wrote the guidance, which was released in late October 2026. It expects lawyers to understand what these tools cannot do and to check every citation themselves. Heidi Brown, an associate dean at New York Law School, says the missing disclosure rule worries her. She also warns of a subtler error that gets less attention.

That to me is more dangerous than just a straight-up hallucination.

Heidi Brown, Associate Dean, New York Law School

Disclosure Is Optional, but the Signature Still Binds

The core instruction is short. A lawyer may use a generative AI tool on a brief or motion and need not tell the court. One thing does not change: the certification that comes with a signature. By putting their name to papers in a New York court, attorneys vouch that nothing in them is false, with the same exposure to sanctions as ever.

The committee presents this as continuity rather than new law. Three duties hold whether a person or a model drafted the text: the duty to be accurate, responsibility for what goes into the papers, and the ban on false statements to the court. Lawyers must grasp a tool's limits. They must also confirm, personally, that nothing it produced is fabricated or wrong.

Chief Administrative Judge Joseph Zayas appointed the committee in 2024. He said the state deliberately chose neither a ban nor a disclosure mandate, because making attorneys answer for their own filings is how the system already works. "Crucially, it reinforces the bedrock principle that attorneys remain fully accountable for the accuracy of the papers they submit to the courts," Zayas said.

The document is a fallback. Where a judge has standing rules on AI, those rules win. It also expands on an approach the court system adopted earlier in 2026. Under that approach, lawyers who choose these tools must understand them and independently confirm that what they produce is reliable.

Why the Committee Says Chatbots Invent Case Law

In legal work, a hallucination is a case, statute or citation that an AI tool presents as real but that does not exist. Judges in New York and across the country have caught filings built on such inventions and sanctioned the lawyers responsible, including in the Southern District of New York.

The committee explains the mechanism plainly. Large language models build an answer word by word, each time choosing the likeliest continuation of what came before. The result is fluent prose, not a retrieval of checked records. "The [AI] is designed to generate text that sounds right, not text that is right," the guidance states. It names ChatGPT, Claude, Gemini and Grok as examples of widely available models. None of them consults a store of verified facts, and none knows whether a citation it supplies matches a real case.

Two further cautions follow. A model knows only its training data, so it cannot reflect developments in the law after a certain date. Training data also carries cultural, economic and social biases, including prejudice against protected classes. Generated text may therefore repeat stereotypes or contain harmful material. The committee's remedy is manual work: locate each authority, read it, and analyse whether it supports the point being made.

Brown's Case for Making Lawyers Disclose

Heidi Brown, an associate dean at New York Law School who helps lead its AI work, called the guidance a good step. She called its silence on disclosure "concerning," as first reported by amNewYork. In her view, it may not stop inaccurate citations or misread precedent from reaching the bench.

"I understand that lawyers have always had a responsibility to check their citations," Brown said. She argued that in a period of mass training on these tools, a disclosure and certification step "might add a layer of realization to lawyers – who maybe aren't as astute about these new tools – to take that extra step to check their work."

Her worry is practical. AI is now built so deeply into legal research platforms that newer lawyers may not realise a passage they filed was machine-generated, or that it carries an AI-induced mistake. A requirement to declare AI use, Brown said, would make an attorney pause over what they had submitted and do the checking the court system says it expects.

Misgroundings Are the Error the Guidance Underplays

Brown's sharper criticism is that the guidance spends so much attention on invented cases that it neglects misgroundings. A misgrounding happens when an AI tool cites a genuine case or statute but gets it wrong. The tool may misdescribe what the authority says, miss the nuance of an argument, or misstate how it connects to other rulings. It may also leave out the latest decision, filing or statutory amendment.

"That to me is more dangerous than just a straight-up hallucination," Brown said. Lawyers must confirm that every proposition, not only quotations and citations, is backed by the sources the AI supplies. She described that as a very time-consuming process.

Newer attorneys find these errors harder to spot, she said. She urged them to become "conscientiously efficient" with the tools rather than treating them as shortcuts. Brown describes herself as pro ethical AI and pro innovation. She still fears these systems are diluting legal writing, and that overloaded judges relying on AI-assisted briefs could let weakened statements of law seep into opinions that then become law.

Will the no-disclosure approach hold up?

The bottom line

New York has chosen trust backed by sanctions over rules backed by paperwork. The court system lets attorneys use generative AI freely and quietly, on the view, set out by Joseph Zayas, that a signature already makes a lawyer answerable for every word filed. Whether that is enough depends on lawyers doing the slow work the committee describes: reading each authority, not just confirming it exists. Heidi Brown's objection lands hardest on that point. A fake case is easy to expose once someone looks. A real case quietly misdescribed is not, and her concern is that such errors could end up in judicial opinions. Judges who want more than the default remain free to demand it.

What to watch

• Watch whether individual New York judges adopt standing AI rules that require disclosure, which would override the default; no timetable has been set.
• Watch for the first sanctions decisions in New York courts that cite this guidance when penalising AI-generated errors; no date has been set.
• Watch whether the advisory committee Joseph Zayas appointed revisits disclosure or adds direction on misgroundings; no review date has been announced.