AI hallucinations trigger 1,490 court sanctions globally
Legal researcher Damien Charlotin has catalogued roughly 1,490 court decisions worldwide involving AI-hallucinated legal material as of May 2026, according to a public sanctions tracker GC AI. The database documents the first comprehensive inventory of judicial penalties imposed on attorneys for submitting fabricated case citations and invented quotations generated by large language models.
More than 1,000 of the recorded sanctions have been issued in the United States GC AI. The penalties range from $5,000 in the seminal *Mata v. Avianca* case to $110,000 imposed by a federal judge in Oregon against two lawyers who filed a brief containing 23 fabricated citations and eight invented quotations GC AI. The Oregon matter was characterized by the presiding judge as "a notorious outlier in both degree and volume" GC AI.
Sanctions Types and Scale
Judicial responses have included fines, attorney disqualifications, license suspensions, pro hac vice revocations, disciplinary referrals, notice orders, and mandatory continuing-education requirements Legal AI Insights. Some jurisdictions have imposed more severe consequences: the Sixth Circuit issued sanctions of $15,000 per attorney, while Nebraska recorded an indefinite bar suspension for hallucination-related misconduct GC AI.
The scale of sanctions accelerated sharply in 2025. During the first two weeks of August alone, three federal courts issued separate sanctions against lawyers for AI hallucinations Agent Times, signaling a shift toward more rapid and consistent judicial enforcement.
Implications for Agent Reliability
The tracker underscores a critical vulnerability in current large language models deployed in high-stakes professional environments. Agents powered by LLMs have repeatedly generated plausible-sounding but entirely fabricated legal precedents, case names, and quotations — output that attorneys relied upon without verification. Courts have responded by treating these errors as sanctions-triggering professional misconduct, not mere computational accidents.
Charlotin's database is maintained as a public resource, allowing legal practitioners and AI developers to review documented cases and sanction outcomes. The compilation reflects only cases that reached judicial determination or public reporting; the actual volume of hallucination incidents in legal filings likely exceeds the recorded sanctions total.
The findings suggest that without structural safeguards — such as mandatory citation verification protocols, AI audit trails in legal documents, or agent output validation frameworks — the integration of AI agents into legal practice will continue to generate disciplinary consequences for attorneys and erosion of professional standards.