Litigation Strategy
When Opposing Counsel Files AI Generated Court Filings
What to do when the motion on your desk cites cases that do not exist, and how to keep your own briefs on the right side of the Wadsworth line.
Legal technology researcher and data scientist specializing in AI governance for litigation teams. Expertise in NLP and AI-assisted document review.
Key Takeaways
- Hallucination sanctions climbed from $5,000 in Mata to $96,000 in Couvrette v. Wisnovsky inside 34 months.
- Section 1927 has no safe harbor and reaches attorneys personally, unlike Rule 11’s 21-day trap.
- Noland v. Land of the Free suggests courts may deny fees to counsel who spot hallucinations and stay silent.
- Verify every citation in every opposing brief and log the time under a dedicated matter code for AI verification from the first minute.
- Default to a footnote, not a sanctions motion, when the filer is a pro se litigant with a sympathetic claim.
The Pro Se AI Flood Reaching Your Docket
Attorney Rudwin Ayala uploaded a draft Motion in Limine to his firm’s in-house AI system and asked it to add case law. The system returned nine citations. Eight did not exist. Walmart’s defense team pulled the brief, spotted the fakes inside an afternoon, and filed for sanctions. Judge Kelly Rankin in the District of Wyoming did the rest, and a Morgan and Morgan partner watched his pro hac vice admission get revoked on the record. That was February 2025, in Wadsworth v. Walmart. April 11, 2026 is where this story picks up, and that’s what opposing counsel’s AI generated court filings can do to your week.
From Mata’s $5,000 to Couvrette’s $96,000
Fourteen months later, the bill got bigger. A federal magistrate in Medford, Oregon ordered San Diego attorney Stephen Brigandi to pay roughly $96,000 in Couvrette v. Wisnovsky, including $80,500 in opposing counsel fees. Local Oregon counsel Tim Murphy drew a separate $14,200 sanction, pushing total exposure above $110,000. That’s the highest reported hallucination-linked sanction as of April 2026, per The Ethics Reporter’s tracking. Arnold and Porter, in a separate Central District of California matter, asked for more than $210,000 just to cover the time chasing fake authority.
If you’re a litigator at the defense table, the question is not whether you’ve seen Mata v. Avianca before. You have. It’s been read to death in CLE panels since 2023. The question is how your practice changes now, when the brief on top of your pile was written by a language model and the person who served it has no supervising partner to check the work. The full sanctions timeline from Mata to Couvrette shows a jump from $5,000 to $96,000 inside 34 months. The curve is not flattening.
The AI Filing Surge By The Numbers
The volume is real. Fisher Phillips, a defense firm with skin in the game, tracked pro se filings jumping from 4,100 to 6,400 in a year. Seyfarth Shaw logged a 40 percent rise in federal pro se ADA Title III filings and 69 percent in Fair Housing Act filings. Bloomberg Law counted 52 court rulings on improper AI use in February 2026 alone. Damien Charlotin’s tracker ran just over 700 cases at the start of 2026 and crossed 1,300 by April.
Here is the sentence that should actually keep you up. Every sanctioned case in this article is a detection success. The honest assumption is that unchecked hallucinations are sitting in briefs on your docket right now, and the only reason you haven’t found them is that neither you nor the judge pulled the cite.
The AI brief does not care whether the filer has a bar card. Wadsworth came from a large plaintiffs’ firm. Couvrette involved a San Diego attorney on pro hac in Oregon, with a client the court said had “persuasively” done the AI drafting herself. Kruse v. Karlen, out of the Missouri Court of Appeals, came from a pro se appellant who paid an online “consultant” a fraction of what a lawyer would charge, and 22 of 24 cited cases were fabricated. Large firm. Solo on pro hac. Pro se appellant. One defensive playbook.
Spotting AI Generated Court Filings Opposing Counsel Sent You
You don’t need a forensic toolkit. You need a reading habit and a Westlaw tab.
Red Flags in AI-Generated Briefs
A handful of signals show up again and again:
- Citations to cases that don’t resolve on Westlaw or Lexis.
- Parentheticals that summarize holdings the real case never reached.
- Reporter volumes that don’t exist or that jumped a decade.
- Writing register that whipsaws from plain English to a polished federal style inside a single paragraph.
- Timelines that can’t be true, such as a seven-page memorandum filed within 30 minutes of the answer (a real example Fisher Phillips partner Steve Cupp described).
- American English from a party whose every pleading has been in British English.
- Stacks of parallel structures in threes.
- Perfectly on-point string cites for an obscure procedural question, the kind only an absurd research budget could produce.
- Fabricated dockets with judges who never sat on that bench.
- Direct quotations that are too neat, too symmetrical, too willing to carry your opponent’s argument.
Spotting is cheap. Verifying is where the hours go, and where the fee recovery record gets built. Run the workflow below mechanically, because a clean contemporaneous ledger under a dedicated matter code is worth several multiples of a reconstructed one when a judge assesses a Section 1927 fee request months later. Build the record from minute one.
The Six-Step Verification Workflow
- Extract every citation from the opposing brief into a tracking sheet. Fifteen to twenty minutes. You need the list before you need the analysis.
- Pull each cite from Westlaw and Lexis on separate tabs. Forty-five to ninety minutes for a typical motion. If a case doesn’t resolve on either service, flag it, do not assume it’s a reporter lag.
- Read the parenthetical against the actual holding for every case that does exist. This is where most AI briefs fall apart. The case is real. The proposition is invented. Sixty to ninety minutes.
- Cross-check the docket and the judge in PACER. Fifteen minutes. Hallucinated judges are rare but devastating when you catch one.
- Log every minute contemporaneously under a dedicated matter code for AI verification work. Courts awarding Section 1927 fees want to see a clean ledger, not a reconstruction.
- Draft a verification affidavit while the facts are fresh, even if you haven’t decided to move for sanctions. You can always not file. You cannot recreate the timeline six weeks later.
Choosing a Citation Verification Tool
A note on the tools. Litigators running this workflow at volume are pairing Westlaw or Lexis with citation verification tools like Hintyr’s, alongside Thomson Reuters CoCounsel Legal and Everlaw’s Writing Assistant. The value is not magic. It’s the ability to surface every proposition, tag the authority behind it, and flag anything that can’t be resolved to a real reporter. Verification still belongs to you. The software just keeps the drudge work from eating your afternoon.
Rule 11, Section 1927, and Fee Recovery for AI Hallucinations
Rule 11 is the rule litigators reach for first, and the rule most likely to cost them.
Rule 11 vs Section 1927: The Trap and the Workhorse
| Dimension | Rule 11(c)(2) | 28 U.S.C. § 1927 |
|---|---|---|
| Safe harbor | 21-day service on opposing counsel before filing | None |
| Reaches | Party and/or counsel | The attorney personally, not the client |
| Mental state | Objective reasonableness | Most circuits: bad faith or recklessness. 11th Cir: “knowingly or recklessly pursuing a frivolous claim” is enough. |
| Scope of review | Paper-by-paper | Course of conduct across the whole litigation |
| Best against | Slow, cautious opponent willing to withdraw | Fast-filing opponent or repeat offender |
| Fee recovery history | Often zero after safe-harbor withdrawal; disciplinary focus | $3,000 (Wadsworth) to $31,100 (Lacey) to $96,000 (Couvrette) |
Rule 11(c)(2) requires you to serve the motion on opposing counsel first, wait 21 days, and only then file it with the court. If you file before the safe harbor expires, you blow the motion. The timing is a trap against a fast-filing opponent: serve, wait 21 days, and your opponent has room to withdraw and walk away clean while your client keeps getting billed.
Section 1927 is the quieter workhorse. It reaches any attorney who unreasonably and vexatiously multiplies proceedings. No safe harbor. The award runs against the attorney personally, not the client. And it looks at a course of conduct across the whole litigation, which matters when the same lawyer sends a second hallucinated brief two weeks after the first. Most federal circuits require bad faith or recklessness, but the Eleventh Circuit holds that “knowingly or recklessly pursuing a frivolous claim constitutes bad faith,” a more litigator-friendly standard worth knowing before you file. In Mata, Judge Castel’s show cause order threatened Rule 11, Section 1927, and inherent authority at the same time. Treat that as the template.
Inherent authority under Chambers v. NASCO, Inc., 501 U.S. 32 (1991), gets you non-monetary sanctions: striking the brief, revoking pro hac vice, disqualification, and referrals to the state bar. Chambers requires a finding of bad faith, which is a higher bar than most Rule 11 motions clear. Johnson v. Dunn, a July 2025 matter out of the Northern District of Alabama, disqualified a Butler Snow team and referred them to the state bar after at least five hallucinated citations came to light. The court said quietly what a lot of judges have been thinking: monetary sanctions aren’t deterring this, and something more is needed.
Federal courts have also layered on a patchwork of standing orders requiring AI disclosure in filings. We covered the full patchwork in our piece on the 300-plus federal court standing orders on AI disclosure, and any sanctions motion you draft should pull the specific order from your judge’s chambers first.
State analogs vary. California CCP 128.7 mirrors Rule 11’s 21-day safe harbor. New York’s 22 NYCRR 130-1.1, invoked in Matter of Samuel, has no such limit. Texas Rule 13 demands a separate good-faith showing and sits in a category of its own. Check your forum.
The fee curve is steepening. Most hallucination sanctions still cluster in the $1,500 to $10,000 range, where the disciplinary component dominates. But the tail is where the money lives. Mata drew $5,000 in 2023. Wadsworth added a $3,000 fine plus pro hac revocation in 2025. Lacey v. State Farm reached $31,100 later that year. Rivera v. Triad Properties hit $47,056.90 in April 2026 in the Northern District of Alabama. Couvrette landed at $96,000 the same month. Arnold and Porter asked for $210,000. The awards keep climbing.
The Noland Signal on Fee Recovery
One move from Noland v. Land of the Free, L.P., out of the California Court of Appeal in September 2025, every defense lawyer needs to internalize. Noland imposed a $10,000 sanction on the filer. The court declined to award fees. Commentators have read Noland as a signal that a practical duty of detection may be forming, one that would flip the economic incentives on opposing counsel. If you see the fabrication and say nothing while the meter runs, the court may decide you forfeited the claim on the fees.
The 14-Day Letter
Here’s the playbook. Draft a short letter to opposing counsel the day you confirm a fake cite. Two paragraphs. Identify the fabrication, cite Mata, demand withdrawal within 14 days. Do not file the letter. Serve it. If opposing counsel withdraws, you’ve preserved your client’s money and built a paper trail. If they don’t, you have contemporaneous notice, which is gold under Section 1927 and inherent authority both. The letter is a filter, not a threat.
Pro Se Judgment Calls and Keeping Your Own Filings Clean
The letter works on counsel with a career to protect. The harder call comes when the filer is a pro se litigant who doesn’t.
Why Most Pro Se Hallucinations Get a Footnote, Not a Motion
You’ve caught a pro se litigant using ChatGPT. Do you move for sanctions? Often the answer is no. Kruse v. Karlen dismissed a pro se appeal and awarded $10,000 to the respondent, but that was an appellate court with 22 fabricated citations out of 24 and a filer who paid an online consultant to do it.
The run-of-the-docket pro se AI user looks different. Usually there’s a sympathetic narrative. Usually the claim has underlying merit. Usually the judge will catch the fake cite on her own, and usually the defendant who moves aggressively for sanctions looks punitive on the record for months afterward. A line of recent decisions, including Wilcox v. Gingrich (Ind. Ct. App. Jan. 2026), is declining to sanction good-faith pro se AI users even where the panel chastised the litigant for nonexistent authorities.
Your first choice on a pro se brief is almost always the footnote. Drop a single sentence in your response noting that the opposing brief cites authorities that cannot be located on Westlaw or Lexis, and attach a short exhibit listing each unverifiable cite. Cite Mata v. Avianca, Inc., 678 F. Supp. 3d 443 (S.D.N.Y. 2023) for the proposition, then argue the merits. Let the court catch it. If the court doesn’t, you’ve still preserved the issue for appeal and kept your own tone measured.
Move for sanctions against a pro se litigant only when three things line up. The fabrication is extensive, not one slip. The merits are weak enough that nothing else stops the bleeding. And your client’s exposure is big enough that the optics of looking aggressive cost less than the money on the table.
Your Own Firm’s AI Policy
Now look in the mirror. If your firm has not written a short internal policy on when associates can and can’t use a language model for research, write it this week. Partners sign every brief. Under Wadsworth, T. Michael Morgan and local counsel Taly Goody each paid $1,000 after signing a motion they had not drafted. Signatures carry duties.
A workflow where a first-year pulls authority from an AI tool, a mid-level cite-checks against the real reporter, and a partner reviews both is defensible under ABA Formal Opinion 512 and supervisory duties under Rules 5.1 and 5.3. A workflow where nobody reads the underlying cases is a Wadsworth waiting to happen. The associate with a browser tab open to a consumer chatbot is not your biggest risk. Your biggest risk is the partner who signs without reading, because that is the pattern every sanctioned firm on this list shared.
The same workflow catches problems before they leave the firm. Tools that pull every proposition and tag its underlying authority (Hintyr is one, Everlaw’s Deep Dive is another in our AI document review tools comparison) run the same verification pass opposing counsel would run against you. The brief goes out cleaner. Your malpractice risk drops. You also take the ChatGPT-in-secret problem off the table for the younger lawyers in your office, because a sanctioned internal tool is less risky than a browser tab nobody talks about.
When the Same Tool Helps a Tenant Win
A Long Beach tenant named Lynn White won her eviction appeal using ChatGPT and Perplexity Pro for $40 a month after losing at trial. She avoided roughly $55,000 in penalties and $18,000 in overdue rent. She told NBC News it felt like David and Goliath, except her slingshot was AI. Her case is real, and so is a growing catalog of legal aid wins powered by the same tools breaking things elsewhere.
Legal Aid Wins the Same Tool Enables
Seventy-four percent of legal aid organizations surveyed in 2025 were using AI, and 88 percent believed it helped close the justice gap. The Legal Services Corporation, which operates on the premise that 92 percent of low-income Americans’ civil legal problems get inadequate or no help, announced $4.2 million in 32 technology initiative grants in December 2025. Courts have started endorsing tools. Legal Aid of North Carolina runs a bilingual voice agent. Nevada’s Administrative Office of the Courts runs a multilingual chatbot. Housing Court Answers in New York City built Roxanne with NYU Law to walk tenants through repair rights.
The Hardest Truth: Identical Docket, Different Intent
Here is the hardest truth in this article. Vexatious AI filings and legitimate pro se AI filings look identical on the docket. The Lynn White brief and the ChatGPT revenge motion are the same genre from the outside. The only way to tell them apart is to read the underlying claim, and that is a burden the profession has not built a fair system to share.
On February 10, 2026, in Warner v. Gilbarco out of the Eastern District of Michigan, a federal magistrate held that a pro se plaintiff’s ChatGPT queries are protected by work product. A small decision with a big implication. Courts are drawing a line that says the tool is legitimate, even as other courts sanction its output. Your job is not to hold both lines. Your job is to verify what’s in front of you, one brief at a time, regardless of who signed it. The ethical duty doesn’t care about the sympathetic plaintiff. The court may, and should. But the duty stays yours.
Get the workflow in place before your name is on the opposition brief. The court won’t wait.
Frequently Asked Questions
What’s the first thing I should do when I suspect an opposing brief contains AI-hallucinated citations?
Extract every citation into a tracking sheet and pull each one from Westlaw and Lexis before you do anything else. Do this the same day you receive the brief, because your fee recovery record starts the minute you notice the problem. Log your time under a dedicated matter code for AI verification from minute one. Draft a short letter to opposing counsel identifying the fabrication and demanding withdrawal within 14 days. Do not file for sanctions before you have tried the letter.
Can I recover fees for time spent verifying opposing counsel’s citations?
Yes, under 28 U.S.C. Section 1927, which reaches attorneys who unreasonably and vexatiously multiply proceedings, with no safe harbor and reaches the attorney personally rather than the client. Courts have awarded fee-shifting sanctions from $3,000 in Wadsworth to $31,100 in Lacey v. State Farm to roughly $96,000 in Couvrette v. Wisnovsky. Contemporaneous time entries under a dedicated AI verification matter code are the single most important factor in the award.
What’s the difference between Rule 11 and Section 1927 for AI hallucination sanctions?
Rule 11(c)(2) requires 21-day service on opposing counsel before filing, which gives the other side a safe harbor to withdraw the brief and walk away clean. It’s a weak tool against fast-filing opponents. Section 1927 has no safe harbor, reaches the attorney personally rather than the client, and looks at a course of conduct across the whole litigation. Most federal circuits require a bad faith or recklessness finding for Section 1927, though the Eleventh Circuit applies a more litigator-friendly standard.
Should I move for sanctions against a pro se litigant whose brief contains fake cases?
Usually no. Pro se AI users often have sympathetic narratives, and courts including the Indiana Court of Appeals in Wilcox v. Gingrich have declined to sanction good-faith mistakes even where the panel called out nonexistent authorities. Drop a footnote in your response identifying the fabricated cite and citing Mata v. Avianca, then argue the merits. Move for sanctions only when the fabrication is extensive, the underlying merits are weak, and your financial exposure justifies the optics.
How can my firm avoid ending up on the Wadsworth side of an AI hallucination case?
Write a one-page internal policy on when associates can and cannot use a language model for research, and require every signed brief to pass a verification step that pulls each citation from a real reporter. Partners sign every brief, and Wadsworth made clear that the partner who signs without reading carries the same exposure as the associate who drafted. Run the same verification pass your opposing counsel would run against you before the brief leaves the firm. ABA Formal Opinion 512 and Rules 5.1 and 5.3 expect exactly this workflow.
Disclaimer: This blog post is published by Hintyr for informational purposes only and does not constitute legal advice. The discussion of ethics rules, procedural rules, and case law is general in nature and may not reflect the rules applicable in your jurisdiction. Attorneys should consult their state bar’s ethics opinions, the local rules of their forum, and qualified legal counsel before making decisions about sanctions motions, fee petitions, or AI-related litigation strategy. No attorney-client relationship is created by reading this post.