Court AI Disclosure Orders: Who Actually Requires Disclosure (2025-2026)

Two layers of court regulation, and only one of them is mapped

When lawyers ask “do courts require AI disclosure,” they are really asking about two different layers. The top layer is system-wide policy: rules adopted by a state court system or supreme court for every courtroom in the jurisdiction. That layer is small, new, and surprisingly permissive. The bottom layer is the judge-by-judge layer: standing orders, individual practice rules, and local rules adopted by single judges or single courts. That layer is large, fragmented, and where all the real disclosure obligations live.

The system-wide layer: permissive by design

New York adopted the first comprehensive statewide rule. 22 NYCRR Part 161, “Use of Artificial Intelligence Technology,” was adopted March 25, 2026 and took effect June 1, 2026, covering every Unified Court System court, civil and criminal. Its posture is accountability over disclosure: AI use in preparing court papers is permitted, and the rule expressly declines to impose a system-wide disclosure requirement. Instead, its Appendix A offers individual courts an optional model rule under which a signature certifies careful review and confirmation that the paper contains no fabricated cases, statutes, or other material, enforceable through 22 NYCRR 130-1.1 sanctions and Rule 3.3. Two scope notes matter in practice: papers offered as evidence are excluded, and because Appendix A is opt-in, New York practice still requires checking whether your particular court adopted it.

Illinois reached the same destination earlier with less machinery. The Illinois Supreme Court’s AI policy, announced December 18, 2024, authorizes AI use by attorneys, judges, litigants, and staff so long as it complies with legal and ethical standards, declines to require disclosure in pleadings, and makes the user fully accountable for the final product, including review of AI content before filing.

New Jersey belongs in this layer too, though its instrument is guidance rather than rule: the Supreme Court’s January 24, 2024 preliminary guidelines state there is no duty to disclose AI use to the court, while underlining that candor duties leave no room for unverified content.

The pattern across all three: courts are not banning the technology or demanding routine disclosure; they are re-pointing existing accountability rules at it. The signature on the filing is the regulatory instrument.

The court-rule layer the tracker maps in full (2025-2026)

Between the permissive statewide policies above and the individual-order layer below sits a third layer: court rules and court-system policies adopted by supreme courts, judicial districts, and appellate courts. In the individual-order layer, Ropes & Gray’s AI court-order tracker categorized 145 orders as requiring disclosure and/or verification when checked on June 10, 2026. This site maps the third layer completely rather than delegating it, and every instrument here is verified against the court’s own published order. The pattern holds with the statewide posture above. Most require certifying the accuracy of what is filed rather than disclosing that AI was used. As of July 2026 the tracker records ten such instruments across eight states:

• New York, statewide, 22 NYCRR Part 161 (effective June 1, 2026): AI use in court papers is permitted with no system-wide disclosure duty; the optional Appendix A turns a signature into a certification of careful review and no fabricated authority.
• Illinois, statewide, Supreme Court Policy on Artificial Intelligence (December 18, 2024): AI use is authorized, disclosure in pleadings is not required, and the user is fully accountable for reviewing AI content before filing.
• Florida, statewide, Rule of General Practice and Judicial Administration 2.515(d)(2) (effective June 15, 2026): a signature represents that the cited legal authorities exist and are accurately cited, an accuracy certification rather than an AI-disclosure rule, with a companion order preempting circuit-level AI requirements.
• Colorado, statewide, Rule Change 2026(02) to the Rules of Professional Conduct (effective January 8, 2026): writes AI into the conduct rules, providing that technology including AI does not diminish a lawyer’s responsibilities, and amending the Rule 1.1 competence comments.
• Arkansas, statewide, Administrative Order 25 plus Rules of Professional Conduct Preamble paragraph 22 and Rule 5.3 (effective December 11, 2025): warns against entering confidential data into public AI tools and confirms that non-human assistance, including AI, does not excuse a violation of the rules.
• Montana, Fourth Judicial District, Missoula and Mineral Counties, Rule of Practice 3.G (October 31, 2025): requires a party using generative AI to disclose the use, identify the tool, explain how it was used, and certify accuracy was checked; a filing without the certification presumptively certifies no AI was used.
• Montana, Thirteenth Judicial District, Yellowstone County, Local Rule 35 (effective January 1, 2026): requires identifying the type of AI used and certifying that every citation to the law or the record has been verified.
• Oklahoma Court of Criminal Appeals, criminal appeals statewide, Rule 1.17 (effective February 18, 2026): every AI-produced or AI-modified portion of a filing must be verified as accurate by a person responsible for the document.
• Rhode Island, statewide, Article V Rule 1.1 comment [6] and Interim Generative AI Guidelines (June 17, 2026): competence now includes AI literacy, and the advisory guidelines ask lawyers to disclose AI use when the client asks or when client information is entered.

Two patterns matter for a filing lawyer. First, only the two Montana district rules impose an affirmative duty to disclose that AI was used; everywhere else the obligation is to certify accuracy, which you owe regardless of the tool. Second, this layer is moving fast, with six of these instruments taking effect in the first half of 2026, so a check that was current last quarter may not be. Each instrument also appears on its state’s deep-dive page with the primary-source link.

The judge-by-judge layer: where the obligations actually are

Individual federal and state judges began issuing AI standing orders in mid-2023, immediately after Mata v. Avianca made fabricated citations a national story, and the layer has grown continuously since. The orders vary widely: some require disclosing any generative AI use in drafting; some require certifying that every citation was human-verified; some apply only to self-represented litigants; a small number prohibit AI-drafted filings entirely.

This site deliberately does not re-catalog that layer, because a definitive free resource already exists: the Ropes & Gray AI Court Order Tracker, a public database of standing orders, local rules, and decisions across state courts, federal district courts, and federal appellate courts, relaunched in May 2026 with judge, court, date, and keyword search. Ropes & Gray’s AI court-order tracker categorized 145 orders as requiring disclosure and/or verification when checked on June 10, 2026; a handful were categorized as prohibiting AI use. Duke Law’s RAILS project also maintains an open dataset, though it has not been updated since May 2025.

What we add instead is the workflow.

The pre-filing check, in four steps

First, check the judge’s individual practices page and standing orders on the court’s own site; that is the authoritative source, and orders change. Second, check the court’s local rules for AI provisions adopted court-wide. Third, run the judge and court through the Ropes & Gray tracker as a cross-check, not a substitute. Fourth, regardless of what you find, run the verification workflow anyway: every citation independently confirmed in a traditional database before filing, because Rule 3.3 and Rule 11-type certifications apply in every courtroom whether or not an AI order exists. A filing that survives that workflow complies with the strictest standing order in the country; one that does not is sanctionable even where no AI order exists. For what that looks like when it goes wrong, our AI sanctions case tracker logs the sanctions decisions we have verified against the courts’ own published orders, with amounts and rule basis.

Firms can make this systematic rather than heroic by writing the forum check into their AI policy as a per-filing step; the policy template includes it, and the candor guide covers what happens to lawyers who skip it.