Litigation Strategy
AI and the Pro Se Flood: How Opposing AI-Generated Filings Change Your Litigation Strategy
The docket used to punish volume. Now it rewards it, and the party on the other side doesn’t bill by the hour.
Legal technology researcher and data scientist specializing in AI governance for litigation teams. Expertise in NLP and AI-assisted document review.
Key Takeaways
- Federal pro se appellate filings hit 20,878 in FY2025, half of all new federal appeals, and ADA Title III filings closed 2025 at 8,667 for the calendar year.
- Chambers inherent authority, not Section 1927, is your lever against a pro se filer in most circuits. The Ninth reaches pro se under §1927; the Third and Seventh have not resolved it. Check your circuit before you draft.
- A show-cause footnote under Rule 11(c)(3) skips the 21-day safe harbor. It’s the right tool when the fabrication is in a brief and you want the judge to do the work.
- Morgan v. V2X (D. Colo. Mar 2026) is one district order, not a rule. Draft your protective order to bar AI upload of confidential discovery unless the provider is contractually forbidden from training on it.
- The vexatious-litigant designation, not sanctions, is what actually stops a serial AI pro se filer. Know the four-factor De Long test in the Ninth Circuit and the five-factor Safir test in the Second.
How AI Filings Turned the Pro Se Flood Into a Capacity Crisis for Defense Counsel
The sibling post covers the sanctions math, the 14-day letter, and the §1927 fee-recovery curve. This one is about the rest of your week, the one where a pro se complaint lands Monday and three more arrive by Friday.
The access-to-justice pitch for generative AI was supposed to cut the other way. Pro se litigants with weak claims would finally get coherent pleadings. That was 2023. The 2026 numbers tell a different story.
The Administrative Office of the U.S. Courts recorded 20,878 pro se appellate filings in FY2025, up nine percent year over year, crossing fifty percent of all new federal appeals. Seyfarth Shaw logged 8,667 ADA Title III filings for the calendar year.
Drafting cost has collapsed. Fisher Phillips partner Evan Shenkman told his firm’s analysts that every pro se plaintiff considers themselves a subject matter expert on the law because they have ChatGPT. His colleague Steve Cupp estimates defendants should plan for a ten-to-fifteen percent spending increase on AI-driven pro se employment matters.
The economics hit defendants fast. Nippon Life Insurance, defending a settled employment suit a former employee reopened with ChatGPT-drafted motions, logged over sixty filings from one plaintiff and is now suing OpenAI to recover roughly $300,000 in defense costs.
Bloomberg Law tracked a sharp rise in federal AI-misconduct rulings across 2025 and into 2026. Q1 2026 sanctions ran into six figures, but verification hours rarely convert into a fee award. You eat it, or your client does.
That is the reversal. Access-to-justice logic still applies to a first-time pro se plaintiff who doesn’t know Twombly from Iqbal. It does not apply to a fifth-time filer cranking out 500-page complaints a commissioner has to read before anyone else gets a hearing.
The First 72 Hours With an AI-Generated Pro Se Complaint
A pro se complaint gets served Monday at 4:47 p.m. You have 21 days under Rule 12(a) to answer or move. Here’s what the first seventy-two hours should look like.
Hour one. Read it. Not the allegations, the structure. A plain-English statement of the parties, a date range that tracks real events, reasonable paragraph pacing. Or the opposite: headings in perfect federal register, three-part parallelism on every page, case names with published parallel cites, and a prayer for relief that stacks declaratory, injunctive, and compensatory. That second shape is your flag.
Hour two through four. Pull every citation into a spreadsheet. Run each through Westlaw and Lexis. Flag what doesn’t resolve. Mark the ones that resolve to a real case but cite a proposition the opinion never reached. The misgrounded cite is the category most likely to sneak past a judge.
Hour five. Check the standing orders. You need to know whether your assigned judge has an AI disclosure rule on the books.
California adopted mandatory Rule 10.430 and advisory Standard 10.80 on September 1, 2025. Rule 10.430 required every trial court to adopt its own AI policy by December 15, 2025, though those policies govern court staff and judicial officers, not party filings. Federal judges are scattered. The standing orders on AI disclosure don’t have national uniformity, and what your district requires is the first thing you confirm before you move.
Hour six. Run a PACER or CourtListener history pull on the plaintiff. Serial filers show up fast. Three or more dismissals on similar theories in the last year means you’re looking at a vexatious-litigant candidate.
Day two. Issue a preservation letter covering chat logs and session exports from whatever generative AI the plaintiff used. You won’t get the logs. You’re building a record.
Day three. Decide whether the fabricated content is in the complaint or in a collateral filing. A 12(b)(6) under Twombly/Iqbal tests implausibility and legal sufficiency. It does not reach the truth of factual allegations, even invented ones. What it does reach are legal conclusions dressed as facts. For hallucinated citations in the complaint or a brief, the right move is a footnote pointing the court to the specific paragraphs and a request that the court issue a show-cause order.
The footnote matters. A sua sponte Rule 11(c)(3) show-cause order bypasses the 21-day safe harbor that slows down every lawyer-initiated Rule 11 motion. The plaintiff still gets notice and a chance to respond, but cannot unilaterally moot the motion by withdrawing the fabrication. That’s procedural advantage you don’t get by filing the Rule 11 motion yourself.
Five Motions That Blunt an AI-Fueled Serial Filer
The five motions below are standard. The order you file them in is not.
1. Motion to Dismiss Under Rule 12(b)(6)
Standard. File it the same week you answer. If the complaint is thin or conclusory under Iqbal, say so and point the court to the precise paragraphs. But don’t cross-examine hallucinated case law in your MTD. Handle that separately. A denial still does work: it forces an amended pleading, and the second bite is a cleaner target for Rule 11 if the same fabrications resurface.
2. Sua Sponte Rule 11(c)(3) Show-Cause Request
A footnote, filed with your opposition or MTD, identifying fabricated citations and asking the court to exercise its sua sponte authority under Rule 11(c)(3). The virtue is the absence of a safe harbor. Rule 11’s 21-day withdrawal window applies only to party-initiated motions. When the court issues the show-cause order on its own motion, the pro se filer cannot cure by withdrawing. Two hours of footnote drafting. Work shifts to chambers.
3. Motion to Designate Plaintiff as a Vexatious Litigant
The heavy motion, and the only one that actually stops a serial filer. De Long v. Hennessey, 912 F.2d 1144 (9th Cir. 1990), sets the Ninth Circuit’s four-factor test: notice and opportunity to be heard, adequate record, substantive frivolousness findings, and a narrowly tailored order. Safir v. U.S. Lines, 792 F.2d 19 (2d Cir. 1986), read alongside Martin-Trigona v. Lavien (2d Cir. 1984), gives you the Second Circuit’s five-factor framework. California’s CCP §391.7 is a statutory pre-filing order plus a statewide Judicial Council list. Texas CPRC §11.054 sets a five-in-seven threshold; §11.101 installs the pre-filing gate. Florida’s 2025 House Bill 1559 expanded its vexatious-litigant statute into family, probate, and small-claims matters effective July 2025.
Pick the right framework. Apply it by the numbers.
4. Motion for Protective Order on AI Upload of Confidential Discovery
Magistrate Judge Dominguez Braswell’s order in Morgan v. V2X, No. 1:25-cv-01991-SKC-MDB (D. Colo. Mar. 30, 2026), barred both parties from uploading confidential discovery to any generative AI tool unless the provider was contractually forbidden from using inputs for training. One ruling, not a rule, but Jefferies v. Harcros Chemicals, No. 2:25-cv-02352 (D. Kan. Mar. 25, 2026), entered comparable language the same month. Your case is stronger with a real reason to ask: medical records with PHI, source code, witness-identifying production. Put the term in your Rule 26(f) proposed order.
5. Targeted Section 1927 Motion Against Counsel (When There Is Counsel)
Save this one for the case where opposing counsel appears. The §1927 circuit split matters here. The statute reaches “any attorney or other person admitted to conduct cases in any court of the United States.” The Second, Fifth, Tenth, and Eleventh Circuits keep pro se parties outside its scope. The Ninth reaches them. The Third and Seventh noted the split without deciding. Check your circuit before you draft. Against a pro se filer where §1927 does not reach, use Chambers v. NASCO, 501 U.S. 32 (1991), inherent authority instead. The sibling post on opposing counsel’s AI generated court filings walks through §1927 and Chambers in detail.
Frame your motion around the court’s time, not your fees. Judges have to read what you cite.
When to Move for Sanctions and When to Just Win the Pro Se Motion
Every week, a defense attorney decides whether to escalate. The rule of thumb the sanctioned-lawyer cases teach is the opposite of what most defense counsel actually do.
Most defense counsel overfile. They see hallucinated cites and reach for Rule 11. Sometimes that’s right. But often it isn’t. Three questions change the answer.
First. Is the plaintiff pro se, and is this the first filing? Dukuray v. Experian declined sanctions in 2024 against a first-offender pro se and issued a forward-looking warning. Judge Kendall in Obi v. Cook County (N.D. Ill. Apr. 2026) took the opposite path, ordering a $5,000 sanction after the plaintiff’s thirteen hallucinated cites followed seventeen from an earlier filing. Kendall called the conduct “egregious, repeated, and ongoing.” The through-line is repetition. First filing gets a warning. Second gets a show-cause order. Third gets money on the record.
Second. Has opposing counsel appeared? Licensed counsel changes the analysis. The signing attorney has Rule 11 duties and §1927 exposure, and the sanctions curve from Mata v. Avianca through Wadsworth and Couvrette is real. Against a pro se, your tools are Chambers inherent authority, the state vexatious-litigant statute, and the sua sponte show-cause path.
Third. What’s the merit of the underlying claim? Wilcox v. Gingrich (Ind. Ct. App. 2026) waived every issue on appeal after finding fourteen fabricated or mischaracterized citations. If the claim is weak on its own merits, win on plausibility grounds and flag the hallucinations in a footnote for the court’s own consideration.
One balancing word. Vexatious-litigant orders and sanctions motions cut both ways. And the same tools that stop a bad-faith serial filer can bury a first-time pro se plaintiff with a real grievance. Reserve the escalation for the repeat offender. The rubric above is built around that.
A coda on your client’s exposure. The 2025-2026 malpractice market bifurcated. Berkley PC issued form 51380, an AI Absolute Exclusion across D&O, E&O, and fiduciary lines. ISO’s CG 40 47 01 26 brought a CGL exclusion effective January 2026. ALPS’s August 2025 guidance warns that intentional-act exclusions may void coverage for attorneys who file AI work without verification. If your client used AI in the underlying drafting, malpractice exposure for a signed AI brief is a separate conversation before the sanctions motion lands.
Building Intake SOPs for the New Normal
Firms without a playbook for this burn out an associate and pay twice. Here’s the intake stack that keeps defense practices out of that hole.
Flag at First Look
Every incoming pleading gets a one-page screen by the intake paralegal. Three checkboxes. Does the filing cite cases? Does the pleading style swing between registers in a single paragraph? Does the plaintiff have a PACER history in the last 180 days on similar claims? Two of three escalates to partner review before the answer is drafted.
Dedicated Verification Time Code
Book it the day the case opens. A discrete matter code, like “AI-VER,” for citation verification hours. Track contemporaneously. Fee awards under §1927 or Chambers collapse without a clean ledger.
Verification Procedure, Not Guesswork
Extract the citations. Run each through Westlaw and Lexis. Read the parentheticals against the holdings. Pull the dockets. Check the judges. Log everything. If you want to speed the first pass, the research process with verifiable pin-cites is the model most verification tools are converging on: extract, check, pin-cite, affirm. The second pass still belongs to a human.
Standing-Order Library
Keep a living table of every AI-disclosure standing order in every district where your firm files. Update it quarterly. Your junior associates do not know which N.D. Cal. judge requires a cover-page certification and which doesn’t. Make that knowledge available to them the day they’re drafting the response.
Preservation Letter Template
Short. Pointed. Demands preservation of all chat logs, session exports, prompt histories, and any artifacts from any generative AI system used in drafting the opposing filing. You send it on day two. You don’t expect compliance. You expect it to frame your later motion.
AI Upload Ban in Every Protective Order
Whether opposing counsel has appeared or not. Borrow the Morgan v. V2X language. Insist on it in the Rule 26(f) meet-and-confer. Your client’s confidentiality obligations require it even where local rules don’t. If a pro se plaintiff uploads your discovery to a public chat system, the breach is on the record and becomes its own sanctions path.
Your Own Drafting Discipline
The exposure runs both ways. Every filing from your firm gets citation verification built into the drafting pass. Noland v. Land of the Free (Cal. Ct. App. 2d Dist. Sept. 2025) put the rule in writing: no paper filed in any court should contain a citation the signing attorney has not personally read and verified.
Frequently Asked Questions
How do I spot an AI generated pro se complaint at intake?
Three fast checks before you open a matter. First, search the caption in PACER or CourtListener for the plaintiff’s name; a serial AI pro se plaintiff shows up within five minutes. Second, pull any two citations from the complaint at random and verify them on Westlaw or Lexis. Third, look for the generative prose tells: uniform paragraph structure, block quotes without pin cites, statutes cited without a jurisdiction. If two of the three hit, price the engagement for a pre-answer motion rather than a merits defense.
Can I get a vexatious-litigant pre-filing injunction against a pro se AI filer who keeps suing my client?
Yes, on a record of repeated frivolous filings. In the Ninth Circuit, De Long v. Hennessey, 912 F.2d 1144, applies a four-factor test: notice and an opportunity to be heard, an adequate record of conduct, substantive findings of frivolousness, and an order narrowly tailored to the abuse. The Second Circuit uses the five-factor Safir v. U.S. Lines standard. California CCP §391.7 and Texas CPRC chapter 11 operate as state-law analogs with statutory thresholds. Document every prior filing, every dismissal, and every warning before you move, and file after the third or fourth frivolous filing, not the first.
What is the Rule 12(b)(6) standard for dismissing claims built on hallucinated cases or invented statutes?
Ashcroft v. Iqbal, 556 U.S. 662, and Bell Atlantic v. Twombly, 550 U.S. 544, do most of the work. A claim propped up by a nonexistent statute fails the plausibility standard because the legal theory cannot exist as pled. Attach a short appendix showing the citation chain and the Westlaw or Lexis result confirming the authority does not exist. Ask for dismissal with prejudice where the fabrications go to essential elements; a denial forces an amended pleading that is a cleaner target for Rule 11 if the same fabrications resurface.
Does the Rule 41 two-dismissal rule work against a serial AI pro se plaintiff who keeps refiling?
Sometimes. Federal Rule of Civil Procedure 41(a)(1)(B) converts a second voluntary dismissal of the same claim into an adjudication on the merits, which carries res judicata weight. Serial AI filers frequently walk away from a losing motion and refile under a slightly different caption, which is exactly the behavior the rule was designed to reach. Plead the history in your answer, preserve the prior dismissal orders, and raise Rule 41(a)(1)(B) the first time the plaintiff tries to nonsuit. Most state two-dismissal rules track the federal version, though timing and scope vary, so check your forum.
How do I stop opposing counsel or a pro se plaintiff from uploading my client’s discovery into a public AI model?
Put an AI-specific clause in the protective order before you produce anything sensitive. Follow the language of Morgan v. V2X, No. 1:25-cv-01991-SKC-MDB (D. Colo. Mar. 30, 2026), which barred uploading confidential discovery to any generative AI tool unless the provider is contractually forbidden from using inputs for training. ABA Formal Opinion 512 reinforces the duty. If a public filing or transcript later shows a party ran your production through a retaining model, move to amend the protective order, request an in-camera audit, and ask for sanctions under Rule 37.