AI Agents for Lawyers

Regulation & Ethics

Court Standing Orders on AI: Where You Must Disclose or Certify

Court AI disclosure rules in 2026: over 30 of 94 US federal districts and 300+ judges require certification, two Ohio judges ban it, Illinois requires none.

By Dr. Niklas SchmidtUpdated 13 November 202610 min read
On this page
  1. Six kinds of order
  2. The Starr lineage
  3. The orders that ask for more than a signature
  4. State courts and the no-disclosure camp
  5. England and Wales, Austria, Germany, arbitration
  6. Evidence rules: the stalled FRE 707
  7. Pre-filing checklist
  8. Sample certification language
  9. Where this leaves you

As of April 2026, more than 30 of the 94 US federal district courts have an AI disclosure or certification requirement, and, counted separately, more than 300 individual federal judges have AI standing orders. Two Ohio judges ban generative AI outright. Illinois says no disclosure is needed in a pleading, and England and Wales have no certification rule at all. The only safe habit is to check the assigned judge's rules before every filing.

Six kinds of order

The two headline counts measure different units. The district figure, from the Drug and Device Law blog and the AI Vortex map, counts courts with a local rule or general order. The judge figure counts individual standing orders and practice rules, and one district can contain a dozen judges with a dozen different wordings. Sorting the orders by what they demand is more useful than either count.

TypeWhat it demandsExamplesTrap
Certify-onlyA certificate that either no part was drafted by generative AI or that AI-drafted language was checked by a human against print reporters or traditional databasesJudge Brantley Starr and Judge Matthew Kacsmaryk (N.D. Tex.); Chief Judge Gilstrap's Local Rule CV-11 (E.D. Tex.); Judge Jernigan's General Order 2023-03 (Bankr. N.D. Tex.)"Traditional legal databases" was written before Westlaw and Lexis embedded generative AI; do not assume the AI layer of a research platform counts
Disclose and certifyDisclose that AI was used, often name the tool, and certify that citations and authorities were checkedJudge Kobayashi (D. Haw.); Magistrate Judge Robertson (E.D. Okla.); Judge Palk (W.D. Okla.); Judges Pratter and Baylson (E.D. Pa.); Judge Padin (D.N.J.)Baylson's order is not limited to generative AI, and Pratter's applies to pro hac vice motions
Disclose tool and method, keep recordsName the tool, describe how it was used, sometimes retain prompts or describe verification stepsMagistrate Judge Fuentes (N.D. Ill.); Judge Martínez-Olguín (N.D. Cal.: lead trial counsel personally verifies and "must maintain records of prompts"); Judge Cronan (S.D.N.Y.: describe "in detail the steps taken to verify"); C.D. Cal. judges Oliver, Slaughter, Blumenfeld and Hwang (separate declaration, naming ChatGPT, Harvey, CoCounsel or Google Bard as examples)A prompt log is discoverable material; keep it clean
Confidentiality attestationCertify that using the tool did not disclose confidential or proprietary information to an unauthorised partyJudge Vaden (Court of International Trade); Chief Judge Hall and Judge Loyd, General Order 23-01 (Bankr. W.D. Okla.)You cannot sign this for a consumer tool with default training
ProhibitionNo AI in the preparation of filings, with carve-outsJudge Newman (S.D. Ohio) and Judge Boyko (N.D. Ohio): exceptions for legal search engines and Microsoft products; W.D.N.C. district-wide order: no AI in research except AI embedded in traditional tools, every citation checked by an attorney or paralegalThe Microsoft carve-out admits Copilot; a Word add-in running another vendor's model does not obviously qualify
Silence with teethNo disclosure, but the signature carries the consequencesJudge Subramanian (S.D.N.Y.): counsel "must at all times personally confirm" AI research; Magistrate Judge van Keulen (N.D. Cal.): hallucinations are imputed to the signature; D. Haw. general order: a "Reliance on Unverified Source" declaration for material not tailored to the caseThese orders look permissive and are the strictest in effect

The primary texts are collected in the Ropes & Gray tracker, the Law360 tracker and a Bloomberg Law comparison table; the UNC Law Library keeps a running bibliography. None is complete, and judges revise their rules without notice, so the tracker is where you start, not where you stop.

The Starr lineage

Judge Brantley Starr of the Northern District of Texas issued the first order in May 2023, as Mata v. Avianca (S.D.N.Y. 2023) was making the problem famous. Its structure has been copied by at least a dozen districts. The filer certifies one of two things: that no portion of the filing was drafted by generative AI, or that any AI-drafted language "was reviewed by a human for accuracy, using print reporters or traditional legal databases". Judge Kacsmaryk's order in the same district is identical in substance, and Texas Opinion 705 points lawyers to the district's Local Rule 7.2(f) as the model of a disclosure rule.

Two features of the Starr wording have aged badly. "Drafted by generative AI" does not say whether a filing whose research was AI-assisted but whose prose was human counts. And "traditional legal databases" predates the point at which every major database put a generative layer on top; a lawyer who verifies a citation by asking CoCounsel or Lexis+ Protégé has arguably not used a "traditional" database at all. The safe reading is the strict one: open the reported case in the database's ordinary search, not its AI assistant, and note that you did.

The orders that ask for more than a signature

Four demands go beyond the standard certificate and catch lawyers who assume every order is the same.

Prompt records. Judge Martínez-Olguín's civil standing order in the Northern District of California requires that counsel "maintain records of prompts and inquiries used in generative AI tools". A firm whose lawyers use a chat interface with history switched off cannot comply. The same problem arises with Judge Cronan's requirement in the Southern District of New York to describe "in detail the steps taken to verify the accuracy of all legal authorities and factual assertions generated by the AI tool"; you can only describe steps you logged.

Separate declarations. Several Central District of California judges require a stand-alone declaration disclosing AI use and certifying that the filer reviewed the source material. It is a distinct document, filed with the brief, and its absence is a procedural defect before anyone reads a citation.

Confidentiality. The Court of International Trade and the Western District of Oklahoma bankruptcy court require a certification that the tool did not disclose confidential or business-proprietary information to an unauthorised party. That is a statement about the vendor's terms and your account settings; the confidentiality checklist shows which tiers of which tools allow you to sign it.

Prohibitions with carve-outs. The Newman and Boyko orders forbid AI in the preparation of any filing but exempt legal search engines and Microsoft products, and Judge Newman's order adds a duty to "immediately inform the Court" on discovering AI use in any filed document. The Microsoft exemption was presumably written with Word's spelling and grammar tools in mind. It now covers Microsoft 365 Copilot, which drafts. Whether a Word add-in from another vendor is a "Microsoft product" is a question no order answers; assume not.

State courts and the no-disclosure camp

The Illinois Supreme Court's policy, effective 1 January 2025, is the clearest statement of the opposite view: AI use "may be expected, should not be discouraged, and is authorized provided it complies with legal and ethical standards", and "disclosure of AI use should not be required in a pleading". Attorneys, judges and self-represented litigants are "accountable for their final work product". Chief Justice Mary Jane Theis said the court's "current rules are sufficient to govern its use". New Jersey is reported to impose no duty to disclose, and secondary commentary describes the Seventh Circuit as the "anti-disclosure circuit". Delaware's interim policy of October 2024 governs judges and staff only.

State trial courts are a patchwork. Judge Peter Weinmann in Erie County, New York, requires a certification that a filing "has been verified and reviewed by a human being". State supreme courts have preferred sanctions to rules: Alabama's dismissed an appeal on 24 April 2026 over AI-tainted briefs and barred the lawyer from filing without a co-signing attorney; New Mexico's imposed $5,000, removal and a bar referral on a lawyer whose murder-appeal brief cited witnesses ChatGPT had invented, with Justice C. Shannon Bacon asking from the bench, as Reuters reported, "Counsel, do you watch the news?" California's legislature passed a lawyer-AI bill on 1 September 2026; whether it has been signed should be checked before relying on it.

One number frames the whole debate. A Northwestern and NYC Bar survey of 112 federal judges, published 27 March 2026, found 61.6% use at least one AI tool, 28.6% use ChatGPT, and 45.5% received no court-provided training. The judges issuing certification orders are, in many cases, users of the same tools.

England and Wales, Austria, Germany, arbitration

England and Wales have no certification rule. What they have is Ayinde v Haringey [2025] EWHC 1383 (Admin), in which the Divisional Court listed the sanctions available under the Hamid jurisdiction, from public admonition and wasted costs to contempt and police referral, and referred the lawyers to their regulators. The judiciary's guidance of 31 October 2025 makes judges responsible for all material in their name; the SRA's Warning Notice of 17 August 2026 sets the "review critically" standard for solicitors; the Upper Tribunal in UK v SSHD [2026] UKUT 81 (IAC) held that pasting client letters into public AI tools puts them in the public domain. The UK guidance page has the detail.

Austria and Germany have no disclosure rules either, and their courts have reached for existing procedure instead. The Austrian Supreme Court in OGH 14 Os 95/25i (7 October 2025) dismissed a defence lawyer's AI-drafted Nichtigkeitsbeschwerde outright under § 285d Abs 1 StPO, without reaching the merits, because it contained invented OGH decisions, and pointed to the lawyer's personal responsibility under § 9 Abs 1 RAO. AG Köln 312 F 130/25 (2 July 2025) found a family-law brief's commentary citations "frei erfunden" and warned of a breach of § 43a Abs 3 BRAO; the Kammergericht Berlin in 17 WF 144/25 (20 November 2025) rebuked counsel for citing a "fantasierende KI". The Austrian and German bar guidance page sets these in context.

Arbitration has no uniform rule in the sources reviewed for this page as of September 2026. Tribunals have case-management power to direct disclosure of AI use, and the practical route is to propose an AI protocol for Procedural Order No. 1 covering verification of authorities, confidentiality of the record and whether any AI-assisted evidence must be flagged. The arbitration practice page has draft language.

Evidence rules: the stalled FRE 707

Proposed Federal Rule of Evidence 707 would have required machine-generated evidence offered without an expert to meet the Rule 702 reliability standard. It was published for comment on 10 June 2025; comments closed on 16 February 2026, with the American Association for Justice and the NYC Bar urging redrafting. It was reportedly not advanced in June 2026 and sent for further study, and the August 2026 to February 2027 comment cycle on the US Courts site lists only Evidence Rules 104 and 902. The earliest effective date is now beyond 1 December 2027. For the moment, AI-generated evidence is handled under existing rules and judicial discretion.

Pre-filing checklist

  1. Identify the assigned judge and read their individual rules and any district general order on AI; record the date you checked.
  2. Classify the order using the table above: certify-only, disclose-and-certify, tool-and-method, confidentiality, prohibition, or silent.
  3. List every AI tool used on the document, including research platforms with generative layers and Word add-ins.
  4. If the order prohibits AI, confirm each tool falls within a carve-out; if not, redraft without it.
  5. Open every cited authority in the database's ordinary search and confirm it exists, says what is claimed, and is quoted accurately; the verification protocol is the full procedure.
  6. Save the verification log to the matter file; if the order requires prompt records or a description of steps, export them now.
  7. Draft the certification in the order's own words, not from a template.
  8. Have the signing lawyer, not an associate, read the certification and the authorities it covers; the casebook shows sanctions attach to the signature.
  9. If a separate declaration is required, file it as its own docket entry.
  10. Diary a check of the judge's rules for the next filing; they change.

Sample certification language

Use these as starting points only. Where an order supplies its own wording, use that verbatim.

Variant A: Starr-style, no generative AI used

CERTIFICATE REGARDING GENERATIVE ARTIFICIAL INTELLIGENCE

The undersigned certifies that no portion of this filing was drafted by
generative artificial intelligence.

Variant B: Starr-style, generative AI used and verified

The undersigned certifies that portions of this filing were drafted with the
assistance of generative artificial intelligence and that all such language,
including every citation to legal authority and to the record, was reviewed by
a human for accuracy against print reporters or traditional legal databases.

Variant C: disclose-and-certify orders

The undersigned discloses that [name of tool] was used in the preparation of
this filing, specifically for [research on / initial drafting of] [section or
pages]. The undersigned personally reviewed the filing, read each cited
authority in full, confirmed that each exists and supports the proposition for
which it is cited, and verified the accuracy of every quotation. The steps taken
were: [list]. This filing complies with Federal Rule of Civil Procedure 11.

Variant C's bracketed list is where Judge Cronan's and Judge Martínez-Olguín's orders bite; write the steps you took, not the steps you wish you had.

Where this leaves you

Court orders on AI are a compliance problem of the ordinary kind: know the rule, follow it, document that you did. The harder problem is the one no order can solve, which is that a certification is only as good as the verification behind it, and the sanctions cases are full of lawyers who signed certificates they had not earned. Litigators drafting with AI should pair this page with the brief and motion drafting guide, and anyone unsure whether disclosure is required in their own jurisdiction can start with the short answer page. Module 6 of the AI Lab for Lawyers course includes a live exercise in reading a standing order and drafting the matching certificate.

Frequently asked questions

Do I have to tell a US federal court that I used AI to draft a brief?

It depends on the district and the judge. More than 30 of 94 districts and over 300 individual judges have orders ranging from a bare certificate that a human verified the content to a requirement to name the tool and describe how it was used. Other judges have nothing. Rule 11 applies everywhere. Check the assigned judge's individual practices before every filing.

Which judges have banned generative AI in filings?

Judge Michael Newman of the Southern District of Ohio and Judge Boyko of the Northern District of Ohio prohibit AI in the preparation of filings, with exceptions for legal search engines such as Westlaw and Lexis and for Microsoft products. The Western District of North Carolina requires a certification that no AI was used in research except AI embedded in traditional research tools.

Does England and Wales require lawyers to certify AI use in court documents?

No. There is no certification or disclosure rule as of September 2026. The courts rely on the existing duties to the court and the sanctions listed in Ayinde v Haringey, from public admonition and wasted costs to regulatory referral and contempt. The judiciary's own guidance makes judges responsible for all material produced in their name.

What should an AI certification in a court filing say?

Match the order. A Starr-style certificate states either that no portion was drafted by generative AI or that any AI-drafted language was checked by a human against print reporters or traditional legal databases. Disclose-and-certify orders also require the tool's name and sometimes the portions affected. Never certify more than you have actually checked; the signature is what the sanctions attach to.

Put this into practice

AI Lab for Lawyers: a live, hands-on course for practising lawyers

4 live sessions (8 hours), recordings included, taught by Dr. Niklas Schmidt of Wolf Theiss. Rated 4.7/5 from 381 ratings; alumni from Norton Rose Fulbright, Paul Hastings, Pillsbury and Taylor Wessing.

Reserve a seat · €200 Maven money-back guarantee · certificate of completion

Questions

Do I Have to Tell the Court I Used AI? Disclosure Rules by Jurisdiction

Whether you must tell the court you used AI depends on the judge: 30+ US federal districts require it, Illinois does not, England has no duty. How to check.

6 min read

Regulation & Ethics

AI Hallucination Sanctions: The Cases Every Lawyer Should Know (2023–2026)

AI hallucination cases reached 2,041 in Charlotin's database by September 2026, 811 involving lawyers. Every sanction from Mata to Couvrette's $110,204, with lessons.

12 min read

Playbooks

How to Verify AI Legal Research: A Citation-Checking Protocol

Verify AI legal research with a five-check protocol: open every authority, match the quote, confirm holding and status, log it. Template, time budgets, tools.

11 min read

Use Cases

Drafting Briefs and Motions with AI Without Ending Up in Charlotin's Database

AI brief drafting is safe only as a five-stage workflow: grounded research, outline, drafting from verified authorities, every case read, court rules checked.

10 min read

Practice Areas

AI for Arbitration Practitioners: Submissions, Evidence and Tribunal Rules

AI for arbitration practitioners: what CIArb's 2025 guideline, tribunal procedural orders and disclosure rules mean for long records, witnesses and quantum.

11 min read

Tools

Microsoft 365 Copilot for Lawyers: Word, Outlook and the Legal Agent

Microsoft 365 Copilot for lawyers as of September 2026. Why 76 percent of firms use it, what $30 a seat buys, the Word Legal Agent, Agent 365 and its limits.

8 min read