Do Lawyers Have to Disclose AI Use to the Court? The 2026 Answer, Layer by Layer

No. There is no nationwide rule that requires a lawyer to disclose AI use to the court. No Federal Rule of Civil Procedure says it, and the two most significant state policies now in force, New York's and Illinois's, both expressly decline to require it. The obligations that do exist live one layer down, in individual judges' standing orders and a patchwork of local rules and administrative orders. And one duty applies everywhere, disclosure rule or not: you verify every citation you sign, because Rule 11 does not care who drafted the brief.

The three layers at a glance

Is there a nationwide rule requiring AI disclosure?

No. No Federal Rule of Civil Procedure, no Federal Rule of Appellate Procedure, and no ABA Model Rule requires a lawyer to tell the court that AI helped prepare a filing. When federal courts punish AI misuse, they reach for tools that predate the technology: Rule 11, 28 U.S.C. § 1927, the court's inherent authority, and the rules of professional conduct.

The absence is deliberate. Both state policies below reason the same way: the verification duty already covers the problem. A fabricated citation violates Rule 11 whether it came from a language model, a sloppy associate, or your own memory. Disclosure rules, where they exist, are individual judges layering transparency on top of that baseline.

What do the 2026 state rules actually say?

Two state court systems have answered the disclosure question at the policy level. Both gave the same answer: disclosure should not be required, because the duty that matters, verification, already binds every filer.

New York: 22 NYCRR Part 161, effective June 1, 2026

The New York State Unified Court System adopted Part 161 effective June 1, 2026, covering every court in the system, civil and criminal. It permits AI use in submissions and answers the disclosure question directly. Its reasoning: the duty to independently review and verify the accuracy of a submission already applies to all papers, AI-assisted or not, so attorneys and parties "should not be required, upon submitting papers, to disclose to the court that they have used AI in the preparation of such papers."

The teeth are in the appendix. Part 161 ships a model rule that any judge may adopt as a part rule: whoever uses an AI tool in preparing a paper must "carefully review the paper and independently ensure that it contains no fabricated or fictitious cases, statutes, or other material," and by signing "certifies that such a review has been conducted and that the paper contains no such fabricated or fictitious content." Per the New York State Bar Association's summary, individual judges remain free to adopt the model rule, write their own, or add nothing. The statewide floor is the existing verification duty. The ceiling is whatever your judge adopts.

Illinois: the Supreme Court's AI policy, effective January 1, 2025

The Illinois Supreme Court announced its AI policy in December 2024, effective January 1, 2025, on the recommendation of the Illinois Judicial Conference's AI task force. It is blunt about permission: AI use by litigants, attorneys, and judges "may be expected, should not be discouraged, and is authorized provided it complies with legal and ethical standards." On disclosure it spends one sentence: "Disclosure of AI use should not be required in a pleading."

The trade matches New York's. Users must thoroughly review AI-generated content before submitting it in any proceeding, and the Rules of Professional Conduct apply in full.

Read together, the policies mark where the state layer is settling: verification, not disclosure. If you owe a disclosure, your state's court system is almost certainly not the source.

Which courts actually require AI disclosure?

The judge-by-judge layer is where the real requirements live. The Legal AI Governance court-order tracker counted 113 active orders binding attorney filings as of spring 2026: 57 in federal district courts, 46 in state courts, the rest in bankruptcy and specialty courts. Of the 113, 82 require verification or disclosure and 9 prohibit AI in filings outright. Treat every number as a floor. Most of these are single-judge standing orders, and judges add, amend, and rescind them continually.

The most sweeping orders so far come from entire circuits. On January 15, 2026, Florida's Eleventh Judicial Circuit (Miami-Dade) issued Administrative Order 26-04: any attorney or self-represented litigant who uses a generative AI tool in preparing a pleading, motion, memorandum, response, or proposed order "must disclose such use on the face of the filing" and certify independent verification. Broward's Seventeenth Circuit followed eleven days later with AO 2026-03-Gen, which reaches further: name the specific tool, and certify that you verified "every citation to the law and/or the record." Both circuits issued coordinated revisions on May 19, 2026, the Eleventh's current order now numbered AO 26-15, keeping the disclosure-plus-certification structure. The Florida Bar's coverage lists the sanctions menu: striking the filing, monetary sanctions, contempt, and bar referral.

What do AI standing orders typically require?

Nearly every order combines two elements:

1. A statement that AI was used. Some orders want it on the face of the filing, some in a separate declaration, and some require naming the specific tool rather than a generic "AI software."
2. A certification of human verification. The signer attests that a person, not the tool, confirmed every citation and factual assertion.

The Eleventh Circuit's required language is the cleanest template of what compliance looks like. It reads, in full:

"Generative artificial intelligence was used in the preparation of this filing. The undersigned certifies that all factual assertions, legal authority, and citations have been independently reviewed and verified for accuracy and accepts full responsibility for the contents of this filing."

The variance is in scope, and scope is where lawyers get caught. Some orders reach only generative AI used for drafting or research; others are broad enough to arguably cover the AI features embedded in Westlaw, Lexis, and Word. Many bind self-represented litigants as well as counsel. A few prohibit AI-prepared filings entirely. Never assume the order in your last case matches this one. Read the actual text, in this court, for this judge.

What happens when you get it wrong?

The sanctions record answers the "does anyone really check?" question.

The anchor is Mata v. Avianca, Inc., 678 F. Supp. 3d 443 (S.D.N.Y. 2023), the opinion that made "ChatGPT lawyer" a phrase. Two attorneys filed an opposition brief citing six judicial decisions that did not exist, then doubled down by filing fake opinion excerpts when the court asked for copies. Judge P. Kevin Castel found bad faith, "acts of conscious avoidance and false and misleading statements to the Court," and on June 22, 2023 imposed a $5,000 penalty jointly on both lawyers and their firm. Note what was absent: no AI standing order existed in that courtroom. Rule 11 and inherent authority were enough.

The record has grown every year since. In May 2025, in Lacey v. State Farm General Insurance Co. in the Central District of California, a special master sanctioned two firms, one among the largest in the country, a combined $31,100 after roughly nine of twenty-seven citations in a ten-page brief proved wrong, at least two pointing to cases that did not exist. Told to correct it, counsel refiled a brief that still contained hallucinated authority, a failure the special master called "tantamount to bad faith." In March 2026 the Sixth Circuit took the record appellate: in Whiting v. City of Athens, a published opinion, it ordered two attorneys whose briefs carried more than two dozen fabricated citations to pay the appellees' attorney fees, double costs, and $15,000 each, with a disciplinary referral. Asked directly whether AI wrote the briefs, counsel never gave the panel a straight answer. Why generated citations fail in exactly this way is covered in our guide to AI citation hallucinations.

The through-line: every sanction on this record punished unverified fabrication, not AI use. No court has sanctioned a lawyer for using AI and verifying the output. Disclosure violations add a count; fabrication is the case.

How do you check whether your judge requires AI disclosure?

Make this a per-case intake step, not a memory exercise:

1. Pull the judge's individual page on the court's website. Every federal district court lists its judges under a "Judges" or "Judges' Info" tab, and each judge's page holds their standing orders and individual practices, usually under exactly those headings. For state courts, start at the circuit's or county's administrative-orders page. This is the authoritative source, and an individual judge's order can override the district's general policy in either direction.
2. Check the district's local rules and general orders. Some districts have adopted court-wide AI rules that bind every case regardless of the assigned judge.
3. Use a tracker as a starting point, never as authority. Trace.law's disclosure-order tracker and Legal AI Governance's court-order tracker both link each entry to the primary source. Confirm on the court's own site before relying on either; orders get amended.
4. Read the order's scope before deciding it doesn't apply. What counts as AI: generative drafting only, or research tools too? Which filings? Disclosure, certification, or both? Does it reach self-represented parties? If the order is ambiguous about the AI baked into your research platform, comply with the broader reading or raise it at the initial conference.
5. Record the answer in the matter file. A one-line note, "Judge X: certification required, all generative AI, all filings," turns a malpractice trap into a checklist item. The check belongs in your firm's written AI policy.

The duty that exists in every courtroom

Strip away the patchwork and one constant remains: Rule 11 means you own every citation in the filing, whether or not AI wrote the draft. Your signature certifies a reasonable inquiry into the law, the same certification it has carried since 1993. Every AI standing order, New York's Part 161, and Illinois's policy all point at that same act: a human confirms that every cited case exists and says what the brief claims it says. The 10-minute verification workflow shows how to run that check before any filing.

That verification is the step worth building tooling around, because it is mechanical and the stakes are not. CaseRead's free Hallucination Shield checks each citation in pasted text (up to 25 per run) for existence and support against real court data, across all 53 US jurisdictions and 9M+ opinions co-sourced with CourtListener, before you sign. To be precise about what that means under a standing order: the Shield supports the verification you certify. The certification itself is yours, and no tool satisfies it for you. And if you want research grounded from the start, CaseRead's answers cite only sources the system actually retrieved — and flag anything that cannot be verified. Plans start free.

The disclosure question will keep moving. The verification answer will not. Check the judge's page before you file, and file nothing a human has not verified.

Frequently asked questions

Do I have to tell the court I used ChatGPT? Only if the specific court or judge you are filing in requires it. There is no nationwide rule, and New York and Illinois both declined to mandate disclosure at the state level. The requirements that exist come from individual judges' standing orders and a small number of court-wide administrative orders, like those in Florida's Eleventh and Seventeenth Judicial Circuits. Check the judge's standing orders before every filing, and verify every citation regardless, because that duty applies everywhere.

Which courts require AI disclosure? There is no complete official list. The Legal AI Governance tracker counted 113 active court orders binding attorney filings as of spring 2026, 57 of them in federal district courts, with 82 requiring verification or disclosure and 9 prohibiting AI in filings outright. Florida's Eleventh and Seventeenth Judicial Circuits require disclosure circuit-wide, from attorneys and self-represented litigants alike. Individual judges add and revise orders continually, so check your judge's page on the court's website in every case.

Does New York require lawyers to disclose AI use? No. 22 NYCRR Part 161, effective June 1, 2026, states that attorneys and parties should not be required to disclose AI use when submitting papers, because the duty to review and verify every submission already applies no matter what tools produced it. Its appendix carries a model rule individual judges may adopt, under which anyone who used an AI tool must carefully review the paper, ensure it contains no fabricated or fictitious cases, statutes, or other material, and certify that review by signing. Check the assigned judge's rules in every case.

What happens if you don't disclose AI use when a court requires it? Violating a standing order or administrative order is sanctionable on its own. Florida's Eleventh Circuit order lists striking the filing, monetary sanctions, contempt proceedings, and referral to the bar. And if the undisclosed AI produced fabricated citations, sanctions follow with or without a disclosure rule: the Mata v. Avianca lawyers paid $5,000 in 2023, two firms in Lacey v. State Farm were ordered to pay $31,100 in 2025, and the Sixth Circuit ordered $15,000 per attorney plus fees and double costs in 2026, all under authority that predates any AI order.

What does a compliant AI certification look like? Florida's Eleventh Judicial Circuit provides the clearest model: a statement on the face of the filing that generative AI was used, plus a certification that all factual assertions, legal authority, and citations have been independently reviewed and verified for accuracy, with the signer accepting full responsibility for the contents. Those two elements, disclosure of use and certification of human verification, appear in some form in most AI standing orders, though the exact wording varies by court.