After ChatGPT-fabricated citations surfaced in a federal filing in 2023, individual judges did not wait for a national rule to emerge. You can read about how courts are sanctioning lawyers for AI-hallucinated citations to understand the punishment record. But well before any sanction is handed down, many judges now require you to clear a procedural hurdle before you even file. U.S. District Judge Brantley Starr of the Northern District of Texas issued what is widely cited as the first standing order on generative AI on May 30, 2023, just weeks after the initial scandal broke. This order required every attorney appearing in his court to file a certificate stating either that no portion of the filing was drafted by generative AI, or that any AI-drafted language had been checked for accuracy by a human being using print reporters or traditional legal databases.
Judge Starr's order quickly became the template for other jurists. Since then, other federal judges and entire court systems have followed with their own versions. By mid-2026, trackers maintained by law firms and legal-AI research groups count well over a hundred separate standing orders, local rules, and policies addressing AI use in court filings. Critically, no two trackers agree on the exact count, and no single federal rule binds every court. This disagreement is the actual story. You are not dealing with a settled body of law, but rather a fragmented, judge-by-judge and court-by-court patchwork.
This guide maps the current landscape of court AI rules. We break down what these rules actually require, show how widely they vary by jurisdiction and even within the same courthouse, look at the live push for a uniform federal rule, and outline practical steps for firms filing in multiple courts.
Judge Starr's order set the template
On May 30, 2023, U.S. District Judge Brantley Starr of the Northern District of Texas entered his "Mandatory Certification Regarding Generative Artificial Intelligence." This order changed the procedural landscape. According to reporting by the law firm Brown Fox Law, the order forced every attorney appearing before him to file a certificate on the docket.
Attorneys had to attest to one of two things. First, they could certify that no portion of the filing was drafted by generative AI. The original order specifically named tools like ChatGPT, Harvey.AI, and Google Bard. Second, they could certify that any language drafted with the help of generative AI was checked for accuracy by a human being. The order required this human verification to be done using print reporters or traditional legal databases.
The enforcement mechanism is simple but severe. The court will strike any filing that lacks the required certificate. Beyond the threat of having a pleading thrown out, an attorney who certifies falsely faces severe professional jeopardy. Falsely certifying accuracy risks Rule 11 sanctions and creates a separate, serious false-statement-to-the-court issue. Judge Starr's model established the prototype. It proved that courts would use a mix of certification, disclosure, and human verification to manage the risks of generative AI. This applies whether you are using general chatbots or specialized AI legal research tools.
Three different things courts are asking for
Court rules do not follow a single standard. When you look across different jurisdictions, the requirements fall into three main categories.
First, courts require certification of human verification. This is the Starr model. It requires an affirmative statement on the docket regardless of whether you used AI. You must declare that you either did not use AI, or that you verified every citation manually.
Second, courts require disclosure of specific tools and methods. Some judges want to know exactly how the technology is used. For example, U.S. District Judge Dale E. Ho of the Southern District of New York has asked attorneys to declare or disclose their use of generative AI in filings. This disclosure must include the specific tool used and a description of how its output was incorporated into the filing. This is a much more granular requirement than a simple yes or no certificate.
Third, courts can outright prohibit generative AI. A tiny minority of courts have banned the technology outright for drafting filings. There is no certification path or verification loophole in these courtrooms. This is the rarest category, with trackers placing the number of outright bans in the single digits or low double digits.
These categories represent materially different obligations. A firm that is set up to provide a simple yes or no certificate might fail the more detailed recordkeeping required to disclose exactly which tool was used and how it was applied.
The lack of a centralized tracking system makes counting these rules difficult. Two well-regarded trackers show different figures:
- The Legal AI Governance Tracker counts 113 active court orders binding attorney filings. It breaks these down into overlapping categories: 73 requiring disclosure, 86 requiring verification, 75 imposing some form of caution, and 10 outright prohibiting AI use.
- The Ropes & Gray AI Tracker counts 147 orders requiring disclosure or verification, and 5 orders that prohibit AI use outright, drawn from a larger pool of 661 cases nationally that have involved AI in some capacity.
This disagreement is not a failure of research. Instead, it reflects how decentralized the court system is. There is no official, court-maintained master list. To add to the complexity, a separate estimate indicates that over 30 federal district courts, meaning the entire district rather than individual judges, have implemented some form of AI disclosure requirement. This makes understanding how accurate legal AI really is a critical part of a firm's risk profile.
It varies by court, by state, and even by judge in the same courthouse
The variations are not just a matter of federal courts versus state courts. The fragmentation runs much deeper.
First, judges within the same federal district often disagree. A tracker dedicated to the Northern District of Texas catalogs standing orders judge by judge. There is no district-wide rule. Requirements differ from one courtroom to the next within the exact same building. A firm with two active matters in the same courthouse could easily face two different AI requirements.
Second, state systems have taken opposite positions on whether disclosure is even a good idea. The Illinois Supreme Court's Policy on Artificial Intelligence, which took effect on January 1, 2025, recommends that Illinois state court judges do not require lawyers to disclose AI use in drafting pleadings. Illinois chose to reject the certification model entirely, showing that the disagreement is philosophical, not just administrative.
Third, some rules that sound like attorney mandates are actually internal operational guidelines. California's Rule of Court 10.430 and Standard of Judicial Administration 10.80 were adopted in July 2025 and took effect in September 2025. This represents the largest state-level AI governance framework in the country. However, this framework governs how California court staff and judicial branch operations use AI internally. It does not dictate what a filing attorney must certify or disclose.
In contrast, other state courts have implemented centralized requirements for practitioners. The New York State Unified Court System adopted a uniform rule on the use of artificial intelligence, which took effect on June 1, 2026. This rule applies across the entire state court system, providing a single standard for New York state practitioners instead of a judge-by-judge patchwork. This is a court-system rule, which is separate from state bar rules. To understand how this compares to your broader ethical duties, read about what the bar actually requires when you use AI.
A national rule may be coming, but it isn't here yet
As of mid-2026, there is still no single Federal Rule of Civil Procedure governing AI disclosure or certification nationwide. Every federal district and individual judge remains free to set their own rules.
This may change in the future. On May 29, 2026, a U.S. Magistrate Judge in the Middle District of Florida formally proposed amending Federal Rule of Civil Procedure 11 to address generative AI use in court pleadings directly. According to analysis by the law firm Barnes & Thornburg, this proposal aims to bring order to the current patchwork.
However, this is only a proposal. It must go through the formal federal rulemaking process, including reviews by the Judicial Conference's rules committees, before it can take effect. This process typically takes well over a year. Until a national rule is formally adopted, you must treat every individual court's order as binding on its own terms.
What this means if you file in more than one court
Practicing in multiple jurisdictions requires active operational management. You cannot rely on a single firm-wide assumption. This is especially true for firms utilizing our guide for solo and small law firms to implement new technologies.
Before filing any document, check the specific judge's standing orders and the court's local rules for AI provisions. Do not assume that a rule from one judge in a district applies to another judge in the same courthouse. The Northern District of Texas proves how much variation can exist under one roof.
You should build a firm-level checklist or intake step for every new matter. Note the specific judge and court, then check for certification, disclosure, or prohibition requirements. Treat this check exactly like you treat judge-specific rules for page limits, margin sizes, or e-filing formats. It is now standard litigation homework.
It is safest to default to the most conservative posture across all your active matters. Certify accurately and disclose the specific tool when a judge's order asks for it. Most importantly, verify every single AI-assisted citation in a primary source. This verification is crucial because your underlying ethical duties of candor and competence apply everywhere. The bar's ethical rules serve as your baseline floor, while court-specific standing orders add docket-specific requirements on top.
Never treat a procedural certification as a substitute for actual verification. Filing the certificate is an administrative task. Checking the cases is the actual safeguard. The sanctions record shows what happens to attorneys who treat certification as a checkbox exercise instead of doing the actual verification.
Finally, if your firm operates across many courts, assign a specific person or a small group to track these rules. These requirements are changing rapidly. Since the first order was entered in 2023, the landscape has shifted constantly. A check you performed six months ago may no longer be accurate today.
FAQ
Is there one federal rule that governs AI use in court filings?
No. As of mid-2026, there is no single Federal Rule of Civil Procedure addressing AI disclosure or certification. A U.S. Magistrate Judge in the Middle District of Florida proposed amending FRCP Rule 11 on May 29, 2026, but that proposal has not been adopted. Until it is, each federal court and individual judge sets its own requirements.
What was the first court order requiring AI disclosure?
U.S. District Judge Brantley Starr of the Northern District of Texas entered the first standing order on May 30, 2023. It required attorneys to certify either that no generative AI was used, or that any AI-drafted language was verified for accuracy by a human using print reporters or traditional databases.
How many courts now have an AI standing order or rule?
Estimates vary because there is no official master list. The Legal AI Governance tracker counts 113 active orders, while the Ropes & Gray tracker counts 147 orders requiring disclosure or verification along with 5 outright bans. Another estimate counts over 30 federal district courts with some form of AI disclosure requirement as of April 2026.
Is a court's AI rule the same as my state bar's ethics rules?
No, they are two separate layers of compliance. State bar rules govern your professional conduct duties everywhere, while a court's standing order is a specific procedural filing requirement. You can comply with your ethical duties and still violate a court's rule if you fail to file a required certificate.
Do all courts require the same thing: disclosure, certification, or a ban?
No. The rules generally fall into three categories, which include certifying verification, disclosing specific tools and methods, or outright bans. Some states have even gone the opposite direction, with Illinois recommending that judges do not require AI disclosure at all.
What should a firm that files in multiple courts do?
Check the individual standing orders and local rules for every judge and court where you file. Build a tracking checklist into your case intake process. Always verify every AI-assisted citation in a primary source, regardless of whether a specific court order commands it.
The bottom line
The procedural rules written by individual courts and judges represent a fast-moving regulatory layer. They are distinct from bar ethics rules and post-hoc sanctions. Because there is no single master list, respected trackers count active orders differently, ranging from 113 to 147. This variance shows how decentralized the legal landscape remains.
The differences are not just a split between federal and state jurisdictions. Requirements vary from judge to judge within the same federal courthouse. Furthermore, some states like New York have established uniform statewide rules, while others like Illinois have explicitly advised judges against requiring AI disclosure.
A uniform federal standard may eventually arrive under a proposed amendment to Rule 11, but it is not yet in force. Until a national rule is formally adopted, multi-jurisdictional firms must examine the standing orders of each individual judge. Relying on verification as a baseline professional duty is the only safe way to navigate this fragmented environment.