A Spoonful Of Sugar | Bryan Sugar's AI Litigation Insights - August 2026

Courts spent the last two years warning lawyers about AI hallucinations. That phase is finished. In the first six days of August alone, seven different courts sanctioned attorneys or litigants for filing AI-generated fabrications, and at least one of those cases included a referral to state bar disciplinary authorities. If you are using generative AI in your IP litigation practice without a rigorous verification protocol, you are no longer only risking embarrassment. You are risking your license.

This Month's Key Takeaway

Courts have moved from education to enforcement. Sanctions for unverified AI-generated filings are escalating in both frequency and severity, and bar referrals are now on the table in IP and other civil matters.

What Happened

In Barteca Holdings v. Tacobarn, decided August 4, 2026, in the District of Connecticut, a lawyer who used three separate AI tools (Open Law, Claude, and ChatGPT) to draft a filing was hit with a $3,500 monetary sanction and a referral to the state bar. The filing contained four false quotations from case law and four misrepresented holdings. This was not a case of a single errant citation slipping through. The court found a pattern of fabricated authority across the submission.
That same week, a lawyer in In re BFI Waste Systems of Tennessee (M.D. Tenn., August 6, 2026) received a public reprimand and $1,500 sanction for similar AI-generated fabrications. In Booker v. U.S. Bank National Association (D. Conn., August 5, 2026), a pro se litigant who consulted four different AI platforms still produced four fabricated citations and multiple false quotations. The tools did not save him. They compounded the problem.

These cases follow the Deutsche Bank v. LeTennier decision from the New York Appellate Division earlier this year, which imposed $5,000 in sanctions for 23 fabricated authorities across five filings and applied a multi-factor framework that courts are now following. The factors include the volume of fabrications, whether the attorney continued after notice, whether corrections were made, the practical impact on the proceedings, and, in some courts, the presence or absence of genuine remorse.

The numbers tell the broader story. The AI Hallucination Cases Database now documents 1,870 cases globally as of August 11, 2026. The Fifth Circuit stated in June that this problem "shows no sign of abating." The Fivehouse decision signaled explicitly that courts "should begin meeting this challenge with an eye towards deterring similar conduct" through "more punitive sanctions."

Why It Matters for IP Litigation

IP litigators face particular exposure for several reasons that are independent of patent law.

First, IP cases are citation heavy. The sheer number of authorities creates more surface area for AI hallucinations to slip through. Modern AI tools often fabricate quotations from real cases or slightly distort holdings. Those errors are harder to catch than an obviously fictitious caption.

Second, IP disputes frequently turn on multi-factor tests and nuanced standards that are especially vulnerable to confident but incorrect AI output. Likelihood of confusion, fair use, substantial similarity, protectability of trade dress, preemption, and trade secret status all involve layered doctrinal elements. When an AI system offers a clean, confident summary that quietly drops or reorders factors, or relies on non-existent authority, it can sound persuasive to a busy practitioner in a specialized area who is focused on the facts and industry context rather than on re-checking familiar black-letter law.

Third, IP litigators often manage cross-jurisdictional portfolios that implicate a patchwork of AI-related orders and rules. There are now 797 court orders addressing AI use in filings, and leading IP venues for trademarks, copyrights, trade secrets, and unfair competition claims, including the Northern District of Texas, the District of New Jersey, and the Northern District of Illinois, have explicit disclosure mandates. Florida adopted a statewide certification requirement effective June 15, 2026, under amended Rule 2.515(d)(2), which requires every signer to certify that any AI-generated content has been verified for accuracy. If you are coordinating brand enforcement, content protection, or trade secret litigation across multiple courts, you must track AI-related obligations alongside local IP rules and standing orders.

Action Items for IP Litigators

1. Implement a mandatory citation-verification step for every filing that involved AI assistance. A human lawyer must independently confirm the existence and accuracy of every cited case, every quotation attributed to a court, and every characterization of a holding in briefs on trademarks, copyrights, trade secrets, false advertising, right of publicity, and related unfair competition claims. Do not rely on the AI tool to verify its own output. Do not rely on a second AI tool to check the first. Obtain the actual opinions from trusted research platforms and read the relevant passages in context.
    
2. Build and maintain a jurisdiction-by-jurisdiction AI obligations matrix for your active IP docket. For each contested matter, track whether the court has a standing order, local rule, individual judge policy, or certification requirement governing AI use. Update the matrix when courts issue new orders, and incorporate that information into your internal checklists for dispositive motions, preliminary injunction papers, and appellate filings.
    
3. Establish a written AI use policy that addresses the "continuing after notice" factor. The Deutsche Bank v. LeTennier framework treats a lawyer who continues to file hallucinated citations after receiving notice far more harshly than one who promptly corrects the record. Your policy should require an immediate review of recent and pending filings whenever a hallucination is identified in any one submission, including related briefs in the same IP matter or in parallel brand, content, or trade secret cases.
    
4. Train your teams on non-obvious AI risks in IP work. Hallucinations are not limited to invented case names. They can appear in fabricated quotations, misdescribed procedural histories, misapplied multi-factor tests, or invented details about how a court treated a particular kind of mark, work, or alleged secret. Include concrete IP examples in training sessions so that associates and in-house counsel can recognize subtle errors in areas where they otherwise feel doctrinally comfortable.
    
5. Document your verification steps in high-risk filings. In matters where you expect close judicial scrutiny, such as emergency motions for injunctive relief in trademark, trade dress, or trade secret disputes, consider memorializing internally which lawyer verified which citations and when. That record can demonstrate good faith if a question arises and may influence how a court evaluates sanctions, particularly on the factors that consider diligence and remorse.

Bryan Sugar is an IP and media litigator at Lewis Brisbois who works at the forefront of lawyers using AI safely and responsibly in legal practice. His practice covers patent, trademark, trade secret, copyright, unfair competition, defamation and right of publicity disputes, and he leverages AI as a tool to deliver better results for clients while reducing litigation costs.

The lesson from this month is clear. AI can be a powerful assistant in IP litigation only when it is paired with the professional judgment and verification discipline that courts now expect.