NYC Lawyer Cited 6 Fake Cases from ChatGPT — Got Sanctioned
Attorney Steven Schwartz used ChatGPT for legal research in a federal case against Avianca Airlines. ChatGPT hallucinated 6 completely fake case citations with real-sounding names, courts, and dates. None existed. The opposing counsel couldn't find them. The judge couldn't find them. Schwartz was sanctioned and fined $5,000. The case became the poster child for AI hallucinations.
The Lawyer Who Trusted ChatGPT: 6 Fake Cases, 1 Real Sanction
What Happened
In early 2023, attorney Steven Schwartz of Levidow, Levidow & Oberman in New York was working on a personal injury case — Mata v. Avianca Airlines. He needed legal precedents to support his arguments.
Instead of using Westlaw or LexisNexis (actual legal research tools), he asked ChatGPT to find relevant case law. ChatGPT obliged — generating 6 case citations that looked completely legitimate. Real-sounding case names. Plausible court names. Reasonable dates. Even fake quotes from fake judicial opinions.
None of the cases existed. Not a single one.
Schwartz filed these citations in a brief to a federal court. When opposing counsel couldn't find the cases, they flagged it. When the judge asked Schwartz to produce the actual decisions, he went back to ChatGPT and asked it to verify. ChatGPT confirmed they were real (they weren't). He even asked ChatGPT if it could provide false cases — it said it could not.
The Fallout
Judge P. Kevin Castel was not amused. In his ruling, he called it an "unprecedented circumstance." Schwartz and his colleague were sanctioned and ordered to pay $5,000 in fines. They also had to notify every judge whose supposed opinion was cited in the fake cases.
The case became international news and the definitive cautionary tale about AI hallucinations in professional settings.
Why This Matters
ChatGPT doesn't know what's true. It generates text that *sounds* right based on patterns. Legal citations are especially dangerous because they follow predictable formats — case name, court, year, volume, page number — so the AI can generate perfectly formatted fake citations with high confidence.
The Lesson
AI hallucinations are most dangerous in domains where they look most plausible. Legal citations, medical dosages, API documentation, financial data — these all follow structured formats that AI can mimic perfectly while being completely wrong.
How to Avoid This
Unlock Full Playbook
Save 3 hours verification of trial and error.
Estimated savings: $5,000+ in sanctions and legal fees
Unlock for $4.99One-time purchase · Instant access · API key included
Steps
- 1Never use ChatGPT or similar tools as a primary research source for factual claims
- 2Cross-reference every AI-generated citation against authoritative databases
- 3Use domain-specific tools for domain-specific research (Westlaw, PubMed, official docs)
- 4When AI provides citations, verify each one individually — don't batch-trust
- 5Train all team members on AI hallucination risks in their specific domain
- 6Implement a verification checklist before submitting any AI-assisted work product
⚠️ Gotchas
ChatGPT will confirm its own hallucinations if you ask it to verify — it doesn't have a truth oracle
Fake citations look MORE real than sloppy real ones — proper formatting ≠ proper facts
Asking AI 'are you sure?' just makes it double down with more confidence
Legal, medical, and financial hallucinations are the most dangerous because they follow structured formats
'I didn't know the AI made it up' is not a defense — you signed the brief, you own it
This applies to EVERY profession, not just law — the lawyer just got caught publicly
Results
Lawyer uses ChatGPT for legal research, files citations in federal court brief
6 fake cases exposed, attorney sanctioned, $5,000 fine, international embarrassment, career-defining mistake
Get via API
Fetch this pitfall programmatically:
curl -X GET "https://api.tokenspy.com/v1/pitfalls/lawyer-fake-cases-chatgpt" \
-H "Authorization: Bearer YOUR_API_KEY"