AI-Fabricated Legal Citations in U.S. Courts (2023)

Incident Summary

Beginning with the landmark case of Mata v. Avianca, Inc. in June 2023, U.S. federal and state courts have sanctioned attorneys in over a dozen cases for submitting briefs containing nonexistent case citations generated by AI tools.^[1]

In the original case, attorney Steven Schwartz used ChatGPT to research legal precedents and filed a brief citing six entirely fabricated cases in the Southern District of New York. Judge P. Kevin Castel imposed a $5,000 fine, finding that Schwartz had acted in “subjective bad faith.” The pattern has since expanded significantly across multiple jurisdictions and law firm sizes.^[2]

Notable subsequent cases include Lacey v. State Farm (C.D. Cal., May 2025), where Ellis George Cipollone O’Brien Annaguey LLP and K&L Gates — the 14th-largest U.S. law firm — were jointly sanctioned $31,100 after an attorney sent AI-generated citations to K&L Gates colleagues who incorporated them into a brief without verification. In Johnson v. Dunn (N.D. Ala., July 2025), attorneys from Butler Snow LLP were publicly reprimanded and disqualified from the case, with the court explicitly stating that monetary sanctions were proving “ineffective at deterring” AI fabrications. By late 2025, a tracking database had identified over 1,000 cases involving AI-fabricated citations nationwide.^[3]

Key Facts

• Scope: Over 1,000 cases identified involving AI-fabricated legal citations across U.S. courts (2023–2025)
• AI tools involved: ChatGPT (OpenAI), Claude (Anthropic), and in-house AI legal research platforms
• Sanctions range: $2,000 (Gauthier v. Goodyear) to default judgment against a client (Flycatcher v. Affable Avenue)
• Firms sanctioned include: K&L Gates (~1,700 attorneys), Morgan & Morgan (42nd-largest U.S. firm), Butler Snow (350+ attorneys)
• Regulatory response: ABA Formal Opinion 512 (July 2024); multiple federal and state courts adopted mandatory AI disclosure requirements
• Escalation: Courts have moved from monetary fines toward disqualification, bar referrals, and default judgment as monetary sanctions proved insufficient

Threat Patterns Involved

Primary: Misinformation and Hallucinated Content — LLMs generate structurally convincing but entirely fictitious legal case citations, complete with docket numbers, judicial names, and fabricated legal reasoning.

Secondary: Overreliance and Automation Bias — Attorneys across firms of all sizes accepted AI-generated citations at face value without applying standard professional verification procedures.

Significance

1. Systemic pattern, not isolated incident — What began as a single embarrassing case has grown into a sector-wide problem affecting law firms of all sizes, from solo practitioners to the 14th-largest firm in the United States, with over 1,000 cases identified by late 2025
2. Escalating judicial response — Courts have progressively increased the severity of sanctions from monetary fines to public reprimand, disqualification, bar referrals, and default judgment, reflecting recognition that financial penalties alone are insufficient to deter the practice
3. Professional infrastructure changes — The ABA’s Formal Opinion 512 (July 2024) established the first professional ethics framework for generative AI use in legal practice, and multiple jurisdictions now require mandatory AI disclosure in court filings
4. Cross-model vulnerability — The problem is not limited to a single AI provider; cases have involved ChatGPT, Claude, and proprietary in-house tools, indicating that hallucinated citations are a structural property of current LLMs rather than a vendor-specific issue