AI Disclosure Requirements for Lawyers: What Courts Require in 2026

A Growing Number of Courts Require AI Disclosure

Attorneys must disclose AI use in court filings across a growing number of jurisdictions. Since 2023, dozens of individual federal judges have issued standing orders on AI use in filings (see the Ropes & Gray AI court order tracker), and a few districts, including the Northern District of Texas, have written AI provisions into their local rules. State courts and legislatures have taken different approaches, covered below. Ignoring a court's disclosure requirement risks sanctions and reputational damage, consequences that far outweigh the minimal effort of compliance.

On May 30, 2023, Judge Brantley Starr of the Northern District of Texas became one of the first federal judges to address generative AI in filings. His judge-specific requirement directed attorneys to file a certificate attesting either that no portion of a filing was drafted by generative AI or that any AI-drafted language was checked for accuracy by a human being using print reporters or traditional legal databases. Judge Starr no longer requires that certificate, because the district's Local Civil Rule 7.2(f) now requires disclosure of generative AI use in briefs.

This guide covers what courts require, what the ABA recommends, and how to build disclosure practices that protect your license. For a broader look at AI ethics obligations, see our ABA rules guide.

The Case That Changed Everything: Mata v. Avianca

On June 22, 2023, Judge P. Kevin Castel of the Southern District of New York imposed a $5,000 penalty, jointly and severally, on attorneys Steven Schwartz and Peter LoDuca and their firm, Levidow, Levidow & Oberman, after they submitted a brief citing six nonexistent cases generated by ChatGPT (Mata v. Avianca, Inc., No. 22-cv-1461 (PKC) (S.D.N.Y. June 22, 2023)). Schwartz had not verified the citations. After the citations were questioned, he asked ChatGPT whether one of the cases was real, and it answered that the case existed.

The sanctions order did not punish the lawyers for using AI. Judge Castel wrote that there is "nothing inherently improper about using a reliable artificial intelligence tool for assistance." He sanctioned them for failing to verify the output and for continuing to stand by the fake cases after they were questioned. This distinction matters: the problem in Mata was not using AI but failing to verify what it produced.

In the wake of Mata, federal judges began issuing standing orders addressing AI-generated filings, and state courts have since adopted their own rules and policies. Trackers such as Law360's federal judge AI order tracker follow the growing list.

Federal Court Disclosure Requirements

Federal courts have addressed AI in filings in three main ways: individual judges' standing orders, district-wide local rules, and reliance on existing rules without an AI-specific mandate.

Standing Orders Requiring Certification

Some individual judges have required attorneys who use AI to disclose that use, certify that they verified it, or both. Examples, including orders since withdrawn or replaced:

• N.D. Texas (Judge Starr): His 2023 order required a certificate attesting either that no portion of a filing was drafted by generative AI or that any AI-drafted language was checked for accuracy by a human being. He no longer requires it, because the district's Local Civil Rule 7.2(f) now requires disclosure of generative AI use in briefs.
• E.D. Pennsylvania (Judge Michael Baylson): A June 6, 2023 standing order requires anyone who used AI in any way to prepare a filing in his cases to disclose that fact and certify that every citation to the law or the record has been verified as accurate.
• S.D. New York (individual judges): Requirements are judge-specific. Judge Dale E. Ho's individual rules, for example, require any party who uses generative AI to generate any portion of a filing to attach a declaration disclosing that use and certifying that the filer reviewed the source material and verified the AI-generated content is accurate. Check each judge's individual practices.
• N.D. Illinois (Magistrate Judge Gabriel A. Fuentes): He issued a standing order on generative AI in 2023, then pulled it back in 2024 after concluding it was no longer necessary and was slightly burdensome. It is a reminder that these orders change.

Local Rules Addressing AI

A few districts have written AI provisions into their local rules, so they apply in every case before every judge:

• N.D. Texas: Local Civil Rule 7.2(f) requires a brief prepared using generative AI to say so on the first page under the heading "Use of Generative Artificial Intelligence." Filing a brief without the disclosure certifies that no part of it was prepared using generative AI.
• E.D. Texas: Local Rule CV-11(g), as amended effective December 1, 2025, keeps all litigants responsible for the accuracy of documents prepared with generative AI and requires them to review and verify any computer-generated content. It is a verification rule, not a disclosure rule.

Elsewhere, including W.D. Texas and M.D. Florida, AI requirements come from individual judges' standing orders, not district-wide local rules.

Courts Without AI-Specific Rules

Many courts have no AI-specific order or rule. There, existing obligations (Model Rule of Professional Conduct 3.3 and Federal Rule of Civil Procedure 11) still apply in full, and using AI does not relax them.

State Court Requirements

State courts are moving fast, and in different directions. California has enacted a statute, effective in 2027, that includes a court disclosure duty. Others, including Florida, New York, and Illinois, have declined to require AI disclosure and rely instead on citation accuracy and attorney verification.

This table is not exhaustive. Check your court's current standing orders and local rules before every filing, because requirements change often.

ABA Formal Opinion 512: Key Rules for AI Use

The ABA has not adopted a standalone AI disclosure rule. Instead, ABA Formal Opinion 512 applies existing Model Rules to generative AI. This guide focuses on four of them. For a deep dive into each rule, see our ethics guide.

Rule 1.1: Competence

Competence now includes understanding how AI tools work, their limitations, and their propensity for hallucination. ABA Formal Opinion 512 (July 29, 2024) says lawyers should either acquire a reasonable understanding of the benefits and risks of the generative AI tools they use or draw on the expertise of others who can advise on those tools' capabilities and limitations. Relying on AI without that understanding risks violating this rule.

Rule 1.6: Confidentiality

ABA Formal Opinion 512 concludes that a client's informed consent is required before inputting information relating to the representation into a self-learning AI tool whose output could lead to disclosure of that information. Attorneys must ensure the AI platform's data handling practices protect confidential information. Opinion 512 does not prescribe a particular product or plan; in practice, many lawyers meet this duty by using commercial plans that do not train on their data by default rather than free tiers that may.

Rule 3.3: Candor Toward the Tribunal

Rule 3.3 prohibits knowingly making false statements of fact or law to a tribunal and requires correcting false statements of material fact or law previously made. A lawyer who files fabricated AI citations and then fails to correct them, as in Mata, risks violating this rule. Separately, Federal Rule of Civil Procedure 11 requires a reasonable inquiry before signing a filing, which means checking every citation. ABA Formal Opinion 512 also tells lawyers to check the court's local rules on AI use and notes that, depending on the facts, noncompliance could raise issues under Model Rule 3.4(c), which bars knowingly disobeying an obligation under the rules of a tribunal.

Rule 5.3: Supervisory Responsibility

ABA Formal Opinion 512 applies the supervisory duties of Rules 5.1 and 5.3 to generative AI: managerial lawyers must establish clear policies on the firm's permissible use of AI, supervisory lawyers must make reasonable efforts to ensure that lawyers and nonlawyers comply with their professional obligations when using it, and the due diligence lawyers owe when outsourcing work to third-party providers also applies to AI providers. This means reviewing, verifying, and taking personal responsibility for every word submitted to a court.

The Heppner Ruling: Work Product and Privilege

In United States v. Heppner, No. 25 Cr. 503 (S.D.N.Y.), Judge Jed S. Rakoff addressed whether a client's own conversations with an AI tool are protected. He ruled from the bench on February 10, 2026 and issued a written opinion on February 17, 2026.

The defendant, facing a criminal investigation, used the consumer version of Claude on his own to research legal questions about his case, then sent the resulting exchanges to his lawyers. The court held those exchanges were neither privileged nor work product, for three reasons:

• Claude is not an attorney. Attorney-client privilege covers communications with a lawyer, and the exchanges were not with one.
• No reasonable expectation of confidentiality. The consumer product's terms and privacy policy allowed the provider to collect and disclose user data, so the communications were not confidential.
• Counsel did not direct the use. The materials were not prepared by or at the behest of his lawyers and did not reflect their strategy, so the work product doctrine did not apply.

Other courts have reached different results on different facts. In Warner v. Gilbarco, Inc., No. 2:24-cv-12333 (E.D. Mich., Feb. 2026), the court protected a pro se plaintiff's ChatGPT materials as work product, reasoning that AI programs are tools, not persons, so using one is not disclosure to an adversary. In Morgan v. V2X, Inc., No. 25-cv-01991 (D. Colo., Mar. 30, 2026) (summary), the court likewise extended work product protection to a pro se plaintiff's AI use, but required him to identify the AI platforms he had used to process material designated confidential under the protective order.

Three practical implications follow:

• Warn clients about consumer AI. A client who runs case questions through a consumer chatbot on their own may be creating discoverable material. Address this at intake.
• Have counsel direct AI use on a matter. The absence of attorney direction is what defeated work product protection in Heppner. Document that AI work on a matter was done by or at the direction of counsel.
• Use tools with confidentiality terms. The confidentiality ground in Heppner turned on consumer terms that permitted disclosure. Commercial and enterprise plans with contractual confidentiality and no training on your data put you in a stronger position.

What to Disclose vs. What You Need Not Disclose

Not every interaction with AI triggers disclosure. There is no uniform line: what counts depends on how your court's order or rule defines AI, and the lists below are a practical starting point, not a rule.

Likely Requires Disclosure Where a Rule Applies

• AI-generated legal research cited in a filing
• AI-drafted sections of briefs or motions
• AI-generated case summaries submitted to a court
• AI analysis presented as the attorney's own legal reasoning

May Not Require Disclosure (Check Your Court's Definition)

• Using AI for grammar and spell-checking, although a broadly worded order (such as one covering AI used "in any way") may reach generative tools
• Using AI to brainstorm case strategy (internal use, not filed)
• Using AI-powered legal research platforms (Westlaw, Lexis+ AI), unless your court's order or rule is broad enough to reach them. Check the definition: N.D. Tex. LR 7.2(f) covers any tool capable of generating new content, whatever it is called
• Using AI to format or organize documents

The line is evolving, and courts define AI differently. When in doubt, disclose: a voluntary disclosure costs little, while ignoring a court's disclosure order can get a filing struck or draw sanctions.

Sample Disclosure Language

Use or adapt these certifications based on your court's requirements:

For Filings Where AI Assisted with Drafting

Certification Regarding Use of Artificial Intelligence

Pursuant to [Court's Standing Order / Local Rule], the undersigned attorney certifies that artificial intelligence tools were used to assist in drafting portions of this [filing type]. Specifically, [Claude / tool name] was used to [describe use, e.g., generate an initial draft of the argument section, summarize deposition testimony]. All AI-generated content has been reviewed, edited, and verified by the undersigned attorney. All legal citations have been independently verified using [Westlaw/Lexis/other]. The undersigned attorney takes full responsibility for the contents of this filing.

Date: ___________
[Attorney Name], [Bar Number]

For Filings Without AI Assistance

Certification Regarding Use of Artificial Intelligence

Pursuant to [Court's Standing Order / Local Rule], the undersigned attorney certifies that no generative artificial intelligence tools, as defined in [Court's Standing Order / Local Rule], were used in the research, drafting, or preparation of this [filing type].

Date: ___________
[Attorney Name], [Bar Number]

Before using this form, confirm that no AI feature in your research or word-processing tools falls within the court's definition. Under N.D. Tex. LR 7.2(f), for example, filing without a disclosure certifies that no part of the brief was prepared using generative AI.

For Filings Where AI Assisted with Research Only

Certification Regarding Use of Artificial Intelligence

Pursuant to [Court's Standing Order / Local Rule], the undersigned attorney certifies that artificial intelligence tools were used to assist with legal research in connection with this [filing type]. All cases, statutes, and legal authorities cited herein have been independently verified by the undersigned attorney using [Westlaw/Lexis/other]. All legal analysis and argumentation reflects the independent judgment of the undersigned attorney. The undersigned attorney takes full responsibility for the contents of this filing.

Date: ___________
[Attorney Name], [Bar Number]

Best Practices: When to Disclose Even If Not Required

Proactive disclosure protects you even in jurisdictions without formal requirements. Consider disclosing when:

• You practice in multiple jurisdictions. A court you appear in tomorrow may adopt a requirement today. Building disclosure into your workflow ensures compliance everywhere.
• The opposing party may challenge your work product. Voluntary disclosure makes it harder for anyone to argue that you concealed AI use.
• Your client agreement addresses AI use. If you promised your client you would disclose AI use, follow through regardless of court requirements.
• The filing contains complex legal analysis. Disclosing AI assistance and your verification process shows thoroughness, not weakness.

Building Firm-Wide Disclosure Practices

Ad hoc disclosure decisions invite inconsistency and risk. Build a firm-wide protocol:

1. Maintain a court-by-court matrix. Track which courts require disclosure, what form they require, and when the requirement was last updated. Assign one attorney to keep it current.
2. Create standard certification templates. Use the language above as starting points. Have templates ready so that adding disclosure takes seconds, not minutes.
3. Log AI use per matter. Record when AI was used, which tool, what prompts were given, and what edits were made. This log shows that AI use on the matter was directed by counsel, the factor missing in Heppner, and supports privilege and work product claims.
4. Train every attorney. Disclosure obligations apply to every attorney who files, not just the ones who use AI directly. A partner who files a brief drafted by an associate using AI must know whether disclosure is required.
5. Review client engagement letters. Update engagement letters to address AI use: whether you will use it, how you will protect confidentiality, and whether you will disclose to courts. Transparency with clients prevents disputes later.