Do You Have to Disclose AI Use in Court Filings? A Judge-by-Judge Framework

Fusion Legal & Tax · November 13, 2026Thought Leadership7 min read

Law firms do not need to choose between using AI and meeting court expectations. They do need a reliable way to identify what each judge requires before a filing leaves the firm.

Federal courts are not converging on a single rule for AI-assisted litigation. They are building a patchwork—one judge, standing order, and set of filing procedures at a time.

That changes the question for law firms. It is no longer enough to ask whether the firm has an AI policy or whether a particular tool is approved. Before anything is filed, the team must also ask: What does this judge require, what conduct triggers the requirement, and who has verified compliance?

The protective response is not necessarily to prohibit AI. It is to build a venue-aware filing gate that preserves the benefits of carefully governed tools while keeping lawyers in control of the representations the firm makes to the court.

Does this judge require an AI disclosure or a certificate?

In a November 2023 Bloomberg Law article, Jessiah Hulle reported on a review of 196 federal court websites that found at least 14 federal courts had released official guidance on AI in litigation. Its central finding was operationally important: “Most courts personalize AI mandates rather than adopt guidelines verbatim,” and “no template has emerged.”

As described in that November 2023 review, the early orders illustrate why a generic “AI was reviewed by a lawyer” policy is not enough:

• Judge Brantley Starr of the Northern District of Texas required a certificate stating either that no generative AI would be used in a filing or that any use would be “checked for accuracy ... by a human being.”
• Judge Stephen Vaden of the Court of International Trade focused on confidentiality, requiring disclosure and certification that generative-AI use did not expose proprietary information to unauthorized parties.
• Magistrate Judge Gabriel Fuentes of the Northern District of Illinois required disclosure but not a separate certificate, relying instead on Federal Rule of Civil Procedure 11 and its requirement that legal arguments be “warranted by existing law.”
• Senior Judge Michael Baylson of the Eastern District of Pennsylvania required disclosure of AI use—generative or otherwise—and certification that citations were “verified as accurate.” As the 2023 review explained, commentators observed that the order’s breadth could reach seemingly routine software such as Grammarly.

These are not stylistic variations. They regulate different events: using a particular category of technology, disclosing that use, protecting information, checking citations, or filing a prescribed certification.

The variation also persisted after the first wave of orders. As reported in July 2025, additional federal judges continued adding AI requirements to their policies and procedures. Colorado U.S. District Judge John L. Kane and Pennsylvania U.S. Magistrate Judge José R. Arteaga added generative-AI certification requirements. Judge Arteaga’s May 15 update required attorneys and self-represented litigants to identify AI-generated portions, disclose the tools used, and certify that citations had been checked for accuracy.

The same 2025 reporting described still other formulations. Judge Kai N. Scott required disclosure of generative-AI use and certification of citation accuracy. Southern District of Texas Chief Judge Randy Crane reminded filers that anyone signing a pleading, motion, or other paper would be responsible for its contents. Judge Marina Garcia Marmolejo similarly stated that, consistent with Rule 11, the signer “remains fully responsible for its content, regardless of whether it was drafted in whole or in part by generative AI.”

For firms litigating nationally—and for Colorado firms appearing before different federal judges—the practical unit of compliance is therefore not simply the court or district. It may be the individual judicial officer and the specific matter.

Why doesn’t tool approval answer the filing question?

Most law-firm AI governance begins upstream: vendor diligence, confidentiality analysis, access controls, acceptable-use rules, and training. Those controls matter, but they do not determine whether a filing needs an AI disclosure or certificate.

That determination happens downstream, at the intersection of the docket and the production workflow. A firm can approve a secure enterprise tool, use it within internal policy, verify the resulting work, and still miss a judge-specific disclosure requirement. Conversely, a team can avoid generative drafting yet overlook an order broad enough to cover another AI-enabled application.

A practical framework should therefore distinguish at least three questions:

1. May the firm use this tool for this information? This is the firmwide governance question.
2. May the litigation team use it for this task? This requires matter-specific analysis of confidentiality, privilege, protective orders, client instructions, and supervision.
3. What must the filer disclose, certify, verify, or avoid? This is the court-facing compliance question.

Collapsing those questions into one “approved/not approved” field creates false simplicity. Tool approval does not answer the filing question.

What should a judge-specific AI order register capture?

A more reliable workflow starts with structured legal research, not institutional memory. At matter opening—and again before a substantive filing—the responsible team should retrieve and record:

• applicable federal and local rules;
• the assigned judge’s standing orders, chambers rules, policies, and procedures;
• magistrate-judge requirements that may apply to referred matters;
• case-specific orders entered on the docket;
• the source URL or docket entry, retrieval date, and effective date; and
• the lawyer responsible for interpreting and updating the requirement.

The register should capture the operative language rather than merely assigning a label such as “AI disclosure required.” That label cannot reliably distinguish among an order covering generative AI, all AI, drafting, research, citations, confidential information, or a particular filing component.

A useful internal record might include separate fields for:

The objective is not to predict what a judge might prefer. It is to translate the actual governing text into a repeatable control.

Who has to verify AI-assisted work before filing?

Human review is meaningful only when the reviewer performs the right checks against authoritative material. A fluent AI output is not evidence that the underlying proposition, quotation, or record reference is accurate.

An August 2025 Thomson Reuters Institute analysis emphasized that lawyers remain responsible for checking facts and citations before filing. Its Westlaw study covering June 30 through August 1, 2025 found 22 matters in which courts or opposing parties identified nonexistent cases in filings, with discipline motions or sanctions following in many instances. The analysis also noted that errors may include inaccurate quotations.

For AI-assisted litigation work, the verification record should therefore show, as applicable, that a lawyer or appropriately supervised team member:

• opened the cited primary authority rather than relying on the generated description;
• confirmed the case name, court, date, reporter or database citation, and precedential status;
• compared every quotation with the source text;
• checked whether the cited page or paragraph supports the proposition stated;
• reviewed subsequent treatment through an appropriate citator;
• matched factual assertions to admissible evidence or the cited record; and
• reviewed the complete filing for internal consistency after revisions.

This does not require attaching an internal research log to every filing. It does require a process capable of supporting any representation the firm makes in a certificate, disclosure, signature block, or response to the court.

What should a lawyer confirm before signing an AI certificate?

A certificate concerning AI use should not be generated automatically from document metadata or added by support staff based on an informal assurance. The certificate is itself a representation to the court.

Before authorizing it, the signing lawyer should know:

• which tools touched the work;
• what those tools did;
• whether another lawyer, staff member, contractor, or vendor used an AI feature;
• which verification steps were completed;
• whether protected information entered any system; and
• whether the certificate’s exact wording is accurate for the entire filing.

That calls for a complete record of how the filing was produced. AI use can occur during research, first drafting, rewriting, citation formatting, document comparison, or proofreading. Asking only whether “the brief was written by AI” is too narrow for orders that regulate portions of a document, particular tools, or any AI-assisted activity.

What should the firm check before every filing?

A law firm’s document-management or filing workflow can turn this patchwork into a manageable process. Before release to the court, the system should require affirmative answers to a short set of matter-specific questions:

1. Were the judge’s current procedures and the docket checked?
2. Did any covered AI system contribute to the filing?
3. If so, is disclosure or certification required?
4. Have citations, quotations, factual assertions, and record references been checked against authoritative sources?
5. Has the signing lawyer reviewed the exact disclosure or certificate?
6. Is the verification record retained with the matter file?

No control can eliminate every drafting or research error. A well-designed filing gate can, however, make responsibility visible, surface judge-specific requirements before submission, and give lawyers a clear opportunity to correct the work.

The broader lesson is that AI competence in litigation is becoming partly procedural competence. Firms need more than approved tools and a thoughtful ethics policy. They need a live map connecting each matter, judge, filing, AI-assisted task, verification step, and required representation.

As federal courts continue to forge their own paths, the firms best positioned to use AI responsibly will not be those that treat every order as a reason to retreat from the technology. They will be those that convert judicial variation into precise, reviewable, lawyer-controlled practice.