New York's Part 161 Makes AI Verification a Court-Wide Duty

What Part 161 requires, stated plainly

Part 161 permits AI, declines a blanket disclosure requirement, and requires the filer to independently confirm no fabricated cases, statutes, or other material. Signing the paper certifies that review happened.

Part 161 permits the use of artificial intelligence in preparing court papers. It does not require attorneys to disclose that they used it. What it does require is a human check: any attorney or party who uses an AI tool in preparing a paper filed in or submitted to a New York court must carefully review that paper and independently ensure it contains no fabricated or fictitious cases, statutes, or other material^[1][2]. The act of signing the paper certifies that this review was conducted.

The rule took effect June 1, 2026, after being adopted on March 25, 2026, and it applies across the Unified Court System rather than court by court^[2]. That system-wide reach is the feature that distinguishes it from the standing-order model that preceded it. Where a standing order binds only the filings in front of one judge, Part 161 sets a single expectation for civil and criminal matters throughout New York's state courts.

The rule also includes an optional model provision, set out in an appendix, that individual courts may adopt. Under that model provision, a signature certifies careful review and confirmation that the paper contains no fabricated cases, statutes, or other material, and it is backed by existing enforcement tools: the sanctions available under 22 NYCRR 130-1.1 and the candor obligations of Rule 3.3^[2]. In other words, the rule does not invent a new penalty regime. It attaches the verification duty to instruments courts already use.

How New York arrived at a verify-first rule

The pressure behind Part 161 is the same pressure that produced individual-judge orders across the country: a documented pattern of filings containing citations that do not correspond to real decisions. Generative systems can produce text shaped exactly like a genuine citation, with a plausible case name, a reporter volume, a page number, and a parenthetical holding, none of which point to an actual case. That output is fluent and confident, which is what makes it dangerous inside a brief and easy to pass through a review focused on form rather than substance.

New York's response was to write one rule for the whole system rather than leave the question to accumulate as a patchwork of individual orders. The Unified Court System adopted the rule on March 25, 2026, with an effective date of June 1, 2026, giving the bar a defined window to adjust before the obligation applied^[2]. The choice to legislate at the system level, rather than to wait for each judge to act, reflects a judgment that the problem was general enough to warrant a general answer.

It is worth being careful about what the rule does and does not settle. Part 161 does not resolve every open question about AI in litigation, and it does not displace the professional-responsibility rules that already govern candor to the court. It adds a specific, signed representation about the contents of a filing, and it makes that representation a court-wide default rather than a courtroom-specific one. The narrower point matters: this is a verification duty attached to a signature, not a broad new code of AI conduct.

The certification hidden inside your signature

Under Part 161 the ordinary signature carries the attestation. Where AI was used, signing represents that a person reviewed the filing and confirmed no fabricated cases, statutes, or other material.

The mechanism Part 161 uses is subtle and easy to underestimate. It does not require a separate certificate or a disclosure box. It makes the ordinary signature on a filing carry an additional meaning: by signing, the filer certifies that the required review was conducted^[1][2]. The signature you already provide becomes the vehicle for the attestation.

That framing changes what a signature represents. Before Part 161, signing a paper carried the familiar representations that attach to a court filing. Under the rule, in matters where AI was used to prepare the paper, the same signature now also represents that a person carefully reviewed the filing and independently confirmed it contains no fabricated cases, statutes, or other material. The representation is present, factual, and testable. A judge, opposing counsel, or a later reviewer can check the authorities line by line, and the signature is what makes any failure the signer's responsibility rather than a diffuse mistake.

The practical consequence is that the gap between believing a filing is clean and certifying that it is becomes a gap you sign across. "I assume the citations are fine" and "I certify a review confirmed no fabricated material" are different statements, and Part 161 puts your name behind the second one. Our companion guide on what court AI-certification orders require you to attest works through the same distinction in the standing-order context, and it applies with equal force here.

Why Part 161 declined a blanket disclosure requirement

A notable feature of Part 161 is what it leaves out. The rule expressly declines to impose a system-wide requirement that attorneys disclose their use of AI^[1][2]. That is a deliberate choice, and it separates New York's approach from the disclosure-first orders that some individual judges adopted earlier.

The reasoning behind a verify-first design is straightforward once stated. A disclosure requirement tells the court that a tool was used, but it does nothing on its own to confirm that the output was checked. A lawyer can disclose AI use and still file a fabricated citation; a lawyer can decline to use AI at all and still file a bad citation copied from a stale memo. Disclosure locates the tool. Verification locates the defect. Part 161 targets the defect directly by requiring the review that catches fabricated material, whether or not that material came from an AI system.

This is not a claim that disclosure is worthless or that New York rejected it on principle. Individual courts may still adopt their own requirements, and other jurisdictions have taken different paths. The point is narrower: New York decided that at the system level, the obligation worth imposing uniformly was the duty to verify, not the duty to announce. For firms, that means the compliance question is not "did we tell the court we used AI" but "can we show that someone confirmed the filing contains no fabricated authorities."

Individual New York courts may still adopt the optional model provision or their own requirements. Part 161 declines a disclosure requirement at the system level; it does not forbid one in a given courtroom^[2].

Illinois reached the same destination a year earlier

New York is not the first court system to choose verification and accountability over mandatory disclosure. The Illinois Supreme Court adopted its Policy on Artificial Intelligence effective January 1, 2025^[3][4]. The two policies were written independently and differ in form, but they land in a similar place on the central question.

The Illinois policy authorizes AI use and states that it should not be discouraged. In its words, the use of AI by litigants, attorneys, judges, and court staff "may be expected, should not be discouraged, and is authorized" provided it complies with legal and ethical standards^[4]. The policy does not require disclosure of AI use in filings. Instead, it holds users fully accountable for their final work product and requires that all users thoroughly review AI-generated content before submitting it in any court proceeding^[3][4]. Attorneys, judges, and self-represented litigants are accountable for what they file, regardless of the technology used.

Read side by side, New York and Illinois describe the same posture in different registers. Both permit AI. Both decline a mandatory disclosure rule. Both put the weight on human review and accountability for the final product. The convergence is the useful signal here: two large court systems, acting a year apart, reached compatible conclusions about where the obligation should sit. That does not make the pattern universal, but it does suggest a direction that firms filing in multiple states should plan around.

How New York, Illinois, and individual-judge orders compare

The three models most litigators encounter in 2026 are New York's system-wide Part 161, the Illinois Supreme Court policy, and the individual-judge standing orders that predate both. They overlap on the core duty and diverge on scope, disclosure, and enforcement. The table below sets out what each one requires, based on the governing texts and, for standing orders, on the aggregate picture that public trackers describe.

Two cautions apply to the standing-order column. First, standing orders are not uniform: their triggers, wording, and consequences differ from one judge to the next, so any single description is a generalization rather than a rule you can rely on for a specific matter. Second, there is no central registry of these orders; the landscape is tracked by law-firm trackers rather than by any official source, and one such tracker reports more than a hundred active court AI orders binding attorney filings^[5]. Treat that count as directional.

Standing-order details vary by judge and are not centrally registered. Confirm the specific order for the judge assigned to your matter before relying on any general description^[5].

What a verify-first regime asks firms to prove

A disclosure rule asks what you said. A verification rule asks what you can show. Part 161 is the second kind, so the defensible answer is a retained record of the review, not a recollection of it.

A disclosure rule and a verification rule ask for different things at the moment of scrutiny. Under a disclosure rule, the question a court can put to you is narrow: did you say whether you used AI. Under a verification rule like Part 161, the question is broader and harder: can you show that the required review actually happened. The first is a statement. The second is a demonstration.

That difference has a direct operational consequence. A verification duty you can defend is one backed by a record of who checked the authorities in a filing, when, and against what. A verification duty you merely hope was satisfied is one you cannot reconstruct if a fabricated citation surfaces later. Part 161 attaches its duty to the signature and its enforcement to existing sanctions and candor rules^[2], which means the useful mental model is not "did I sign" but "can I show the review if a court asks." The signature is the visible part. A retained verification record is what makes it defensible.

This is where the verify-first design rewards firms that treat verification as a producible artifact rather than a private habit. Our guide on AI hallucinations in legal filings covers why fabricated authorities slip through in the first place, and our overview of AI legal citation checkers compares the tooling that can run the underlying check. The rule does not tell you to buy any tool. It tells you the review must happen and, by signing, that you are representing it did.

Where verification tends to break down under deadline

The duty Part 161 imposes is not complicated to describe, yet it fails in predictable ways. The failures are rarely a lawyer deciding not to check. They are the check being done unevenly under deadline pressure, or done once and then not repeated after a late edit introduces a new citation, or done by someone who confirms a case exists but not that it still holds what it is cited for. The failure modes are procedural, not moral, which is why a signed attestation alone does not prevent them.

Two checks are skipped more often than the rest. The first is confirming that a proposition is actually supported by the authority cited for it, not merely that the authority is real. A genuine case cited for something it does not hold is still a defective citation, and it is exactly the kind of defect a fluent AI draft can introduce. The second is confirming that the authority remains good law, because a case that was correct when decided may have been reversed, vacated, or superseded since. Both require going past the surface of the citation, which is the step that gets compressed when time is short.

Part 161 does not distinguish between a citation that was fabricated outright and one that was real but mis-cited; the duty is to confirm the filing contains no fabricated or fictitious cases, statutes, or other material, and to review it carefully^[1]. In practice, that means the review has to catch both kinds of defect before the signature goes on the paper. A record that captures who checked these items and when is what lets a firm demonstrate the work was done rather than assumed.

A late edit that adds or changes a citation after the review pass is a common source of failure. Tie verification to the moment citations are finalized, and run it again on anything added afterward.

Building an evidence record that survives a challenge

The through-line of a verify-first rule is evidence. If the obligation is to confirm a filing contains no fabricated authorities, and the enforcement is existing sanctions and candor rules, then the practical question is whether you can produce a reproducible check when a filing is questioned. This is the problem RankShield Legal is built to address. It certifies that cited authorities exist, are quoted accurately, and are good law, and it produces a verifiable record you can retain alongside a filing. It is a vendor tool that supports the verification duty; it is not legal advice and does not replace attorney judgment.

Mapped to Part 161, the "no fabricated cases or statutes" requirement is addressed by authority verification that confirms each cited decision is real; the accuracy of a citation by checking the reporter, volume, and page against the actual authority; the fidelity of a quotation by comparing the quoted language to the source; and the good-law question by validation against subsequent history that flags reversal, vacatur, or supersession. The output is an independently checkable record that every cited authority was resolved against live case law before filing. Our page on citation certification describes what that certificate contains and how a reviewer inspects it.

Two honesty points belong here, because this is where vendors tend to overreach. First, no tool makes a filing "hallucination-free," and none should claim to; what a check provides is an independent, reproducible pass that catches fabricated and mis-cited authorities before they reach a judge, plus a record you can produce. Second, the duty stays with the signer. Verification produces evidence that supports the attestation Part 161 attaches to your signature; it does not discharge it. The judgment about whether an authority genuinely supports the proposition it is cited for remains an attorney's to make.

1. Confirm existenceVerify that every cited case, statute, or authority is real, which is the core of the Part 161 "no fabricated or fictitious material" duty.
2. Confirm accuracy and supportCheck that the citation is accurate, that any quotation is verbatim, and that the authority actually holds what it is cited for.
3. Confirm good lawValidate each authority against subsequent history to catch reversal, vacatur, or supersession before the paper is signed.
4. Retain the recordKeep a verifiable record of who checked what and when, so the review behind your signature can be shown if the filing is questioned.

What this means for firms filing across state lines

A firm with matters in New York, Illinois, and various federal courts now faces three overlapping regimes at once: a system-wide New York rule, a statewide Illinois policy, and a shifting set of individual-judge standing orders in the federal system with no central registry. The disclosure triggers differ, the enforcement language differs, and the scope differs. Memorizing every variant is not a workable strategy, and assuming last quarter's process still fits this filing in this court is how firms get caught out.

The more durable answer is to build the process around what every one of these regimes is ultimately about: whether the authorities in a filing are real, accurately cited, and still good law. New York asks you to confirm no fabricated material^[1]. Illinois asks you to thoroughly review AI content and holds you accountable for the result^[3][4]. Individual orders ask, in varying words, for verification and sometimes disclosure^[5]. A single verification step that runs on every cited authority satisfies the substance of all three, while a per-court check on the specific disclosure and wording requirements handles the parts that vary.

Built that way, the workflow does not depend on anyone remembering to be careful under deadline. It reads the governing rule or order for the court and judge, runs an independent check on every cited authority for existence, accuracy, quotation, support, and good-law status, records who verified and when, and retains a record that can be produced if the filing is questioned. That approach meets Part 161 today, aligns with the Illinois policy, and covers the standing-order obligations that apply in the meantime, without betting on any one regime staying still.

This article is informational and reflects the state of these rules as of July 2026. It is not legal advice. The author is RankShield Legal's founder, not an attorney. Consult a licensed attorney about the obligations that apply to your matter, and confirm the current text of any rule or order before relying on it.