Unmasking the Lawyers Behind AI’s Legal Blunders and Why Small Practices Are Bearing the Brunt
- Small Firms Dominate the Drama: Over 90% of AI hallucination cases involve solo practitioners or firms with fewer than 25 lawyers, highlighting how resource-strapped attorneys are most vulnerable to these tech pitfalls.
- Plaintiffs Lead the Pack: In 56% of incidents, it’s the plaintiff’s counsel submitting flawed filings, compared to just 31% for the defense, revealing uneven risks in adversarial battles.
- ChatGPT Takes the Crown: Among tools named in these mishaps, ChatGPT appears in half the cases, underscoring the dangers of relying on popular but imperfect AI for high-stakes legal research.
In the high-stakes world of litigation, where every citation can make or break a case, artificial intelligence was supposed to be a game-changer—a tireless assistant sifting through mountains of case law to deliver precise, reliable insights. Instead, it’s become a notorious troublemaker, churning out “hallucinations”—fabricated cases, bogus quotes, or outdated references that have left lawyers sanctioned, clients underserved, and judges exasperated. Despite court orders banning unchecked AI use, ethics opinions from bodies like the American Bar Association, and a flood of continuing legal education (CLE) courses warning of the risks, these blunders keep piling up. A landmark case in June 2023, Mata v. Avianca, first exposed the issue when a lawyer cited nonexistent precedents generated by ChatGPT, sparking a wave of scrutiny. But who’s really behind these filings? Drawing from a comprehensive database of AI hallucination cases maintained by French lawyer and scholar Damien Charlotin, this analysis dives into 114 U.S. court incidents from June 2023 to October 7, 2025, painting a broader picture of how AI is reshaping—or rather, rattling—the legal landscape.
Charlotin’s database is a goldmine for understanding these errors, categorizing them by type: outright fabricated cases, twisted quotes from real ones, or invocations of overturned precedents. It doesn’t just tally mishaps; it spotlights the humans caught in the crossfire. For this deep dive, the data was pulled as a CSV file on October 9, 2025, and filtered to focus solely on U.S. cases involving lawyers, federal defenders, or paralegals—excluding pro se litigants to zero in on professional accountability. After cleaning for duplicates and misclassifications, 114 cases remained, sourced from court orders, dockets, and firm websites. Firm sizes were binned into categories like solo, 2-25 attorneys, and larger brackets, with solos treated separately to acknowledge their unique pressures. While the sample isn’t exhaustive—new cases emerged almost immediately, including updates and fresh pro se matters—it offers a snapshot of a growing epidemic, one that evolves faster than any single analysis can capture.
One of the most striking patterns emerges in the party dynamics: plaintiffs are far more likely to submit these tainted filings. In 56.1% of the cases (64 out of 114), the blame fell on plaintiff’s counsel, versus 30.7% (35 cases) for the defense. The remaining 13.2% (15 cases) involved “other” scenarios, like bankruptcy proceedings, family law disputes, probate matters, tax court filings, agency actions, habeas petitions, or even an attorney disciplinary hearing where AI errors surfaced in the lawyer’s own defense. For appellate cases, classifications mirrored the trial-level posture to maintain consistency. This imbalance isn’t random; plaintiffs often bear the burden of initiating claims with robust, precedent-heavy arguments, where AI’s promise of quick research might tempt hurried filings. Defendants, by contrast, may lean more on established defenses or have deeper resources for verification. Yet, in adversarial arenas from civil suits to criminal defenses, these hallucinations ripple outward, wasting opposing counsel’s time chasing ghosts and burdening courts already strained by caseloads.
Even more revealing is the firm-size breakdown, which underscores how AI’s pitfalls disproportionately ensnare the little guys. Across the 114 cases, 129 firms or entities were implicated (counting multiples where co-counsel from different outfits shared blame, but only faulted ones). Solos made up a whopping 50.4%, while small firms of 2-25 lawyers accounted for another 39.5%—together, over 90% of the total. Larger players were rarities: just 3.1% from 26-100 attorney firms (like Ellis George or Hagens Berman Sobol Shapiro), 2.3% from 201-500 (such as Butler Snow or Goldberg Segalla), and a mere 1.6% from mega-firms exceeding 1,000 lawyers (K&L Gates and Morgan & Morgan). No firms in the 701-1,000 range appeared, and government attorneys popped up in only two instances—a public defender and county counsel. This skew toward solos and small shops isn’t surprising in a profession where 75% of U.S. lawyers work in practices under 10 people, per industry stats. These attorneys juggle heavy caseloads without the luxury of research teams or robust fact-checking protocols, making AI’s siren call of efficiency all the more irresistible.
Repeat offenders paint an even starker portrait of vulnerability. Five lawyers appeared in multiple cases, all from solo or small-firm backgrounds: Maren Miller Bam of Salus Law and solo Tyrone Blackburn (each in two cases), Jane Watson of Watson & Norris (admitted to the bar in 2024), Chris Kachouroff of McSweeney Cynkar & Kachouroff (infamous for a Zoom mishap), and family-court attorney William Panichi, who faced scrutiny in four matters over a single 30-day span—prompting him to wind down his practice and surrender his license. These stories humanize the data: they’re not faceless errors but the fallout of overworked professionals grasping at tech lifelines.
When the database specified tools— in just 30% of cases (34 total)—ChatGPT dominated, implicated in half (18). Westlaw’s AI features trailed, followed by Anthropic’s Claude, Microsoft Copilot, Google Gemini, and LexisNexis tools. Some attorneys juggled multiple AIs, amplifying the risks. This reliance on ChatGPT, often via free or in-house versions like the Ghostwriter Legal app, speaks to its accessibility for budget-conscious solos. Yet, as a recent Stanford study co-authored by my colleagues revealed, even premium legal AIs like GPT-4, Lexis+ AI, Westlaw’s AI-Assisted Research, and Ask Practical Law AI hallucinate at alarming rates—far from the “hallucination-free” hype they peddle. These tools’ shortcomings aren’t abstract; they’re eroding trust in the profession.
At its core, this crisis exposes deeper fault lines in legal practice. Small and solo firms, under pressure from billable hours, client demands, and personal strains like health issues or caregiving, turn to AI as a “godsend,” as one 404 Media report aptly described. The legal industry faces immense pressure to adopt these technologies, yet they falter precisely where accuracy is paramount. The consequences extend beyond embarrassed attorneys facing sanctions: clients in high-stakes arenas like criminal cases or parental rights battles suffer subpar representation; opponents burn hours debunking fictions; and courts grapple with policing ethics violations while fending off bogus precedents from seeping into opinions. Pro se litigants, absent from this analysis, fare worse—comprising 160 cases in the full database, a majority in the U.S., where unaffordable representation drives self-filing and betrayal by cheap AI tools.
So, what now? The ABA’s July 2024 opinion on generative AI hasn’t stemmed the tide; dozens more cases have surfaced since. To curb this, courts at every level—federal, state, tribal, agency; civil, criminal, or otherwise—might mandate declarations from appearing lawyers affirming AI’s fallibility and staff training. State bars could enforce AI-specific CLE, akin to mandates on substance abuse or bias elimination. Even then, slip-ups will persist; the mantra must be “trust, but verify.” AI developers bear responsibility too: their marketing must match reality, with robust disclaimers overblown claims. As tools evolve (as users demand), so must education—but perfection, from human or machine, remains elusive.
These 114 cases illuminate a profession at a crossroads: AI could bridge the gap for small practices craving reliable research, yet its current flaws widen inequities. Until tools deliver on their promise without the hallucinations, they risk becoming not saviors, but saboteurs—piling unnecessary work on lawyers, courts, and the justice system they serve. The lesson? In law, as in life, cutting corners with unvetted tech invites disaster. It’s time for the bar to demand better—from itself and the innovators vying for its business.