AI Standing Orders in Court: What Filers Must Now Certify

Some judges will not read your brief until you have filed a separate piece of paper swearing to how the brief was made. The question on that paper is narrow and specific: did a generative AI tool draft any part of this filing, and if so, did a human being check what it produced. This is the new world of the AI standing order in court, and it pays to understand where it came from and what it asks before you appear in front of a judge who has one.

Where this started

The first of these orders came from Judge Brantley Starr of the U.S. District Court for the Northern District of Texas, issued on May 30, 2023. His standing order created a Mandatory Certification Regarding Generative Artificial Intelligence. It requires every attorney and pro se litigant appearing before him to file a certificate attesting to one of two things: either that no portion of any filing was drafted by generative artificial intelligence, or that any language drafted by generative AI was checked for accuracy by a human being before submission. The order specifies the method of checking, namely the use of print reporters or traditional legal databases, and it names ChatGPT, Harvey.AI, and Google Bard as examples of the tools it covers. Judge Starr was blunt about his reasoning, observing that while these platforms are powerful and have many uses in the law, legal briefing is not one of them, because the systems are prone to inventing quotations and citations.

That order became a template. Since 2023, the requirement of a judge AI disclosure has spread well beyond one courtroom in Texas. By most accounts, hundreds of state and federal judges have since issued standing orders, general orders, local rules, or pretrial orders addressing the use of AI in filings. Several organizations now maintain running trackers of these orders, including Law360 Pulse and the Responsible AI in Legal Services (RAILS) project. For a firm with a multistate or multi-judge docket, that fragmentation is the practical problem: there is no single national rule, and the orders are not uniform.

What the certification actually asks

It helps to be precise about what these orders do and do not demand, because lawyers often overestimate the burden. The typical AI certification filing is not a confession and it is not a prohibition. Most orders, following the Starr model, give you a clean choice. You may certify that no generative AI was used, or you may certify that it was used and that a human verified every AI-assisted assertion, including citations and quotations, against authoritative sources.

The variations among judges fall into a few recognizable patterns:

• Certify-or-verify orders. The most common form. You attest either to non-use or to human verification of any AI output. Judge Starr's order is the archetype.
• Disclosure-of-extent orders. Some judges ask you to describe how and where AI was used in the document, not merely that it was checked.
• Flat prohibitions for certain materials. A smaller set of orders bar reliance on generative AI for legal research or drafting altogether in matters before that judge, or require leave of court.
• No special order, existing duties only. Many judges have declined to add anything, taking the position that current rules already require accuracy.

The thread running through all of these is verification by a human. Even an order that permits AI use conditions that permission on a person, not a machine, confirming that the cited cases exist, say what the brief claims they say, and remain good law. That is not a novel obligation. It is the same diligence the rules of civil procedure have always required of the signer of a pleading.

Why judges did this

The orders did not appear in a vacuum. They were a direct response to filings that contained authorities the machine had simply invented.

The case that prompted much of this is Mata v. Avianca, Inc., 678 F. Supp. 3d 443 (S.D.N.Y. 2023). Two attorneys submitted an opposition brief that cited and quoted six judicial decisions that did not exist, among them a fictitious Varghese v. China Southern Airlines. One of the lawyers had used ChatGPT to find favorable authority and then asked the chatbot whether the cases were real, accepting its assurance that they were. On June 22, 2023, Judge P. Kevin Castel imposed a $5,000 sanction under Federal Rule of Civil Procedure 11 on attorneys Peter LoDuca and Steven A. Schwartz and their firm, Levidow, Levidow & Oberman. The court's central finding is the one worth remembering: the problem was not the use of a tool, but the abandonment of the lawyers' gatekeeping responsibilities after the tool produced its output.

The pattern repeated at the appellate level. In Park v. Kim, 91 F.4th 610 (2d Cir. 2024), decided January 30, 2024, the Second Circuit confronted a reply brief that cited a nonexistent decision the attorney admitted she had generated with ChatGPT and had been unable to produce when asked. The court referred the lawyer to its Grievance Panel, explaining that the use of such tools does not relieve an attorney of the independent duty to ensure that submissions are accurate and legally tenable. These are not isolated episodes. Trackers of AI-related sanction decisions now list a steady stream of similar incidents across trial and appellate courts, which is precisely why the standing-order movement has continued rather than faded.

A patchwork, not a national rule

It would be convenient if there were one authoritative rule to follow. There is not, and the most instructive event on that front is a decision not to act.

In November 2023, the U.S. Court of Appeals for the Fifth Circuit proposed amending its rules to add an AI certification requirement, which would have obligated counsel to certify either that no generative AI program was used or that any AI-generated text, including citations and legal analysis, was reviewed for accuracy by a human. After a public comment period drew substantial opposition, the Fifth Circuit announced on June 12, 2024 that it would not adopt the proposed amendment. The court reasoned that existing rules already make parties and their counsel responsible for ensuring the accuracy and truthfulness of their filings, so a dedicated AI rule was unnecessary. Critics had also called the proposal ambiguous and warned it could chill legitimate uses of the technology.

The takeaway for a managing partner is that the landscape is genuinely mixed. One federal appellate court concluded that a special AI standing order court rule added nothing to duties lawyers already owe, while individual district judges across the country have concluded the opposite and imposed certification requirements case by case. Both can be true at once, and both can govern different matters on your firm's docket in the same week. The only safe operating assumption is that the requirement might apply, and that you will not know unless you check.

What the ethics rules require regardless

Even where no judge has entered an order, the professional conduct rules already supply the substance of what the orders formalize. The American Bar Association addressed this directly in Formal Opinion 512, "Generative Artificial Intelligence Tools," issued July 29, 2024, its first formal ethics guidance on the subject. The opinion maps the use of these tools onto familiar Model Rules:

• Competence (Rule 1.1). Lawyers must understand, at least at a reasonable level, the benefits and limitations of the specific tools they use, and must keep that understanding current.
• Confidentiality (Rule 1.6). Lawyers must understand how a tool uses input data and must guard against disclosing client information. The opinion cautions that boilerplate consent buried in an engagement letter is generally not sufficient when client confidences will be put into a tool that learns from or retains them.
• Candor to the tribunal (Rule 3.3) and meritorious claims (Rule 3.1). Lawyers must guard against AI hallucinations becoming the basis for frivolous arguments or false statements to a court.
• Supervision (Rules 5.1 and 5.3). Partners and managers must adopt measures giving reasonable assurance that lawyers and nonlawyer staff using these tools comply with the rules.
• Fees (Rule 1.5). A lawyer generally may not bill a client for time spent learning a tool for general use, though the analysis differs where a client specifically requests a particular tool for the matter.

Read together, Opinion 512 and the standing orders point in the same direction. A certification that a human verified AI output is simply a written promise to do what candor and competence already require. If your firm's practices satisfy the ethics rules, signing the certificate should be a formality rather than a scramble.

A practical compliance approach for small firms

For a firm of three to thirty lawyers, the goal is a light, repeatable process that makes any required certification true and easy to sign. A few concrete steps:

• Check for an order before every appearance in a new court or before a new judge. Make this a docketing step. Standing orders typically live on the judge's page on the court website, and the public trackers can be a useful cross-check.
• Keep the firm's standard certification language in a template. When an order applies, you want to fill in a known form, not draft one under deadline.
• Adopt a written AI-use policy. State which tools are approved, what client information may never be entered into a general consumer tool, and who must verify any AI-assisted research.
• Require independent citation verification. Every cited authority should be pulled and read in a traditional reporter or established legal database, and every quotation confirmed against the source. This is the single discipline that would have prevented Mata and Park.
• Document the human review. A short note in the file recording who checked the authorities and when turns an abstract duty into evidence of diligence.
• Train supervised staff, not just partners. Rules 5.1 and 5.3 make supervisors answerable for associates, paralegals, and contract attorneys alike.

The defensible posture is not to swear off these tools, and it is not to use them freely and hope. It is to use them with a verification step that you can attest to honestly when a judge asks. The certification is, in the end, only as good as the human review behind it, and that review has been the lawyer's job all along.

This is general information for lawyers and law-firm leaders, not legal advice, and it does not create an attorney-client relationship. The authorities are cited so you can read them yourself.