AI Disclosure Templates for Court Filings (12 PDFs)

On April 4, 2026, Magistrate Judge Mark Clarke of the District of Oregon ordered San Diego attorney Stephen Brigandi to pay $109,700 for signing a brief with fifteen fake citations and eight fabricated quotations. Brigandi’s own client, a serial pro se litigant, had drafted the brief with AI. After the show-cause order landed, Brigandi produced a 2018 letter, written in 2025, arguing he’d disclosed AI use for years. Judge Clarke applied a per-item approach: $500 a citation, $1,000 a quotation.

And on March 31 alone, seventeen other U.S. courts flagged suspected hallucinations in a single day. If your firm’s AI disclosure paragraph is still the one someone pasted from a 2023 standing order, this post is for you.

Per ABA Journal reporting, the $109,700 order imposes about $96,000 on Brigandi himself (a $15,500 sanction plus his share of an $80,500 opposing-fee award), with the balance falling on his local counsel. Three weeks earlier, on March 13, 2026, the Sixth Circuit issued Whiting v. City of Athens, TN (Stranch, Bush, Murphy, JJ.), hitting Van R. Irion and Russ Egli with $15,000 per attorney, joint and several liability for opposing counsel’s full fees, plus double costs, for more than two dozen fake or misrepresented citations across three consolidated appeals. The court grounded the sanction in FRAP 38 and inherent authority under Chambers v. NASCO. Not § 1927.

Add those two orders to the rest of the quarter and you get more than $145,000 in publicly reported sanctions against lawyers for AI hallucinations in a single quarter. Damien Charlotin’s AI Hallucination Cases Database logged 1,330 cases worldwide and 901 in the U.S. as of April 17, 2026. For the full jurisdiction tracker, see the 300+ court rules tracker we publish. For how this sits alongside malpractice risk, the pattern’s blunt: courts writing these opinions no longer treat hallucinations as first-offense mistakes.

You can’t file one paragraph into every court and call it a day. The rules disagree, sometimes loudly. A filing that satisfies Judge Brantley Starr in the Northern District of Texas can run afoul of the Illinois Supreme Court’s 2025 policy, which tells lawyers the opposite thing.

Start with the vocabulary problem. A certification says what you did: you verified every citation, you checked every quotation, you supervised any nonlawyer tools. A disclosure says what you used: you tell the court you ran a draft through a specific AI product. Some judges want both. Many want only the certification. A few want neither. Conflating the two is how firms end up over-disclosing work product they didn’t have to reveal.

Federal district court has three main archetypes. Judge Brantley Starr (N.D. Tex.) issued the first mandatory AI certification in May 2023, warning lawyers that platforms are “prone to hallucinations and bias” and lack “duty, honor, or justice.” He’ll strike any filing missing the certificate. Judge Michael Baylson (E.D. Pa.) took a broader cut on June 6, 2023, covering any AI use and requiring disclosure plus verification of every citation. Judge Stephen Vaden at the U.S. Court of International Trade added a confidentiality layer on June 8, 2023: counsel must swear no confidential or proprietary information was disclosed to a generative model.

And judges move. Magistrate Judge Gabriel Fuentes in the Northern District of Illinois rescinded his AI standing order in August 2025, calling it “no longer necessary and slightly burdensome.” The Northern District of Texas went the other way and adopted Local Rule 7.2(f), effective September 2, 2025, which treats the absence of disclosure as a negative certification by operation of the rule. The Fifth Circuit rejected proposed Rule 32.3 in June 2024 with a warning that “‘I used AI’ will not be an excuse for an otherwise sanctionable offense.”

State court pushes back in the other direction. The Illinois Supreme Court AI Policy, effective January 1, 2025, says flatly: “Disclosure of AI use should not be required in a pleading.” The New Jersey Supreme Court’s Preliminary Guidelines (January 25, 2024) agree. Over-disclose in those courts and you’re volunteering strategy you didn’t owe anyone.

The templates are starting points. These four rules keep you from turning a prudent certification into an accidental work-product waiver or a sanctionable omission.

Rule 1. Certify verification, don’t volunteer disclosure.

ABA Formal Opinion 512 (July 29, 2024) anchors the rule. It tells you that before submitting materials to a court, lawyers must “review these outputs, including analysis and citations to authority,” and folds the duty into Model Rules 1.1, 1.4, 1.5, 1.6, 3.1, 3.3, 5.1, 5.3, and 8.4(c). Your certification should attest to that verification. It shouldn’t describe which product you used unless the local rule or standing order demands it. Judge Vaden’s confidentiality certification shows you can warrant that no privileged material touched a model without naming the model. If you want to see why product-level disclosure can blow back, read Rakoff’s February 2026 privilege holding.

Rule 2. Every signing attorney verifies, not just the drafter.

This is the lesson of Wadsworth. The Morgan & Morgan brief was drafted by an associate using the firm’s internal AI platform. The senior signatories didn’t personally check the citations. Judge Rankin sanctioned all three anyway. T. Michael Morgan and Taly Goody each paid $1,000 because signature means verification. Your certification should restate that rule in plain English: every lawyer who signs warrants that they, personally, pulled each cited case. Delegation to the drafter isn’t delegation of the duty. See our breakdown of Rule 5.1 and 5.3 supervision duties for how this maps onto partner review pipelines.

Rule 3. Carve out classical retrieval, not generative output.

Pull every generative feature back into the verification promise, regardless of vendor. The Western District of North Carolina standing order excludes Westlaw, Lexis, Bloomberg Law, and Fastcase, but that carve-out covers Boolean search and traditional headnotes, not the newer generative layers. Westlaw Precision AI, Lexis+ AI, and CoCounsel can still produce the same fabricated cites as a consumer chatbot. T1 and T5 draw the line exactly where the courts are drawing it: classical retrieval gets the carve-out, generative output stays inside the verification attestation. Read the tool documentation before you check any box.

Rule 4. Know when not to file these at all.

The Illinois Supreme Court AI Policy says disclosure “should not be required in a pleading.” New Jersey’s Preliminary Guidelines agree. File a long AI certification into an Illinois circuit court that didn’t ask for one and you’ve just told opposing counsel what your research pipeline looks like. You’ve also handed them an interrogatory hook into your firm’s workflow. File certifications where they’re required or prudent. File nothing beyond verification attestations where the state supreme court has said plainly that they aren’t needed.

Brigandi’s backdated 2018 letter is what a firm invents when verification happened nowhere else. These four rules aren’t about disclosure for its own sake. They’re about having a real record before a show-cause order arrives.

This post is for general information about AI disclosure rules and sanctions trends. It isn’t legal advice and doesn’t create an attorney-client relationship. The templates are starting points, not court-approved forms. Before filing, confirm the current local rules, your assigned judge’s standing orders, and applicable ethics opinions, and use your own professional judgment.