AI HALLUCINATION EXAMPLES

Legal AI Citation Hallucination Examples

Real-world examples of AI-generated citations that look authentic but are completely fabricated. See the patterns and learn how to spot them.

No client facts needed. These are illustrative examples only.

Understanding AI Citation Hallucinations

Large language models generate text by predicting the most likely next word based on patterns in their training data. When asked to cite legal authorities, they often produce citations that:

  • Look authentic: Follow proper Bluebook formatting with correct punctuation
  • Reference real concepts: Use real legal terms, court names, and jurisdiction abbreviations
  • Seem plausible: Combine elements in ways that appear reasonable to non-experts
  • Are completely fabricated: Point to non-existent cases, statutes, or reporters

The examples below demonstrate these patterns. None of the "fake" citations exist in real legal databases — but they all look like they could be real to someone unfamiliar with legal research.

Important: Verification Required

These examples are for educational purposes only. Always verify every AI-generated citation against authoritative legal databases (Westlaw, Lexis, Bloomberg Law, official court sites) before relying on it.

Category 1: Fake Case Citations

Vanderbilt v. Tennessee Valley Authority, 999 F.4th 1234 (6th Cir. 2024)

Why it's a hallucination:

  • Reporter issue: "F.4th" is not a real federal reporter series (F.3d is current for 6th Circuit)
  • Reporter number: 999 is impossibly high for any F. series reporter
  • TVA reference: Tennessee Valley Authority is real, but this case doesn't exist
  • Plausibility: Follows perfect Bluebook format, making it look authentic

CiteClear detection: Flags as suspicious due to impossible reporter number and format

Smith v. Imaginary Corporation, 1234 U.S. 567 (2023)

Why it's a hallucination:

  • Reporter number: 1234 is impossibly high for U.S. Reports (Supreme Court cases)
  • Party name: "Imaginary Corporation" is a giveaway, but AI often uses real-sounding names
  • Year: 2023 is plausible, but Supreme Court doesn't decide 1234 cases per year
  • Format: Otherwise perfect Bluebook citation format

CiteClear detection: Flags as suspicious/malformed due to impossibly high reporter number

In re: AI Generation Litigation, 888 F.Supp.3d 999 (D. Fantasy 2025)

Why it's a hallucination:

  • Court: "D. Fantasy" — no District of Fantasy exists
  • Reporter: 888 F.Supp.3d is very high (but potentially real)
  • Year: 2025 may be in the future when generated
  • Case name: "In re: AI Generation Litigation" sounds real but doesn't exist

CiteClear detection: Flags as suspicious due to non-existent court

Category 2: Fake Statute Citations

42 U.S.C. § 19830

Why it's a hallucination:

  • Section number: Section 19830 of Title 42 doesn't exist (real § 1983 is civil rights)
  • Plausible: Title 42 and section format are correct
  • Pattern: AI often adds extra digits to real statute numbers

CiteClear detection: Cannot verify existence without database access, but flags as needing review

28 U.S.C. § 12526

Why it's a hallucination:

  • Section number: § 12526 doesn't exist in Title 28 (Judiciary and Judicial Procedure)
  • Real sections: § 1251-1260 cover appellate jurisdiction, but not 12526
  • Pattern: AI extends real section numbers with extra digits

CiteClear detection: Flags as needing manual verification

18 U.S.C. § 1030(a)(9)

Why it's a hallucination:

  • Subsection: 18 U.S.C. § 1030 (computer fraud) has subsections (a)(1)-(7), but not (9)
  • Plausible: Format is correct, and § 1030 is a real statute
  • Pattern: AI often invents non-existent subsections of real statutes

CiteClear detection: May not flag this automatically — manual verification required

Category 3: Fake Regulation Citations

42 C.F.R. § 483.999

Why it's a hallucination:

  • Section number: § 483.999 doesn't exist in 42 C.F.R. (Public Health)
  • Real range: Part 483 covers Medicare/Medicaid requirements, sections go up to ~483.75
  • Pattern: AI often uses .999 or similar high numbers

CiteClear detection: Flags as suspicious due to pattern

29 C.F.R. § 1910.10000

Why it's a hallucination:

  • Section number: § 1910.10000 doesn't exist in OSHA regulations
  • Real pattern: Part 1910 covers occupational safety, with sections like 1910.1020 (access to exposure/medical records)
  • Format: Otherwise correct CFR citation format

CiteClear detection: Flags due to suspiciously long section number

Category 4: Mixed/Plausible Examples

Some AI-generated citations are harder to identify as fake because they follow all the right patterns:

Johnson v. Tech Innovations LLC, 945 F.3d 123 (9th Cir. 2020)

Why it might be a hallucination:

  • No obvious errors: Format is perfect, reporter number (945) is plausible for F.3d
  • Court is real: 9th Circuit exists
  • Year is plausible: 2020 is within range
  • Problem: This exact case doesn't appear to exist in legal databases

CiteClear detection: Cannot prove non-existence — requires manual database verification

Lesson: Even perfectly formatted citations must be verified

Doe v. Metropolitan Transportation Authority, 876 F.Supp.2d 456 (S.D.N.Y. 2019)

Why it might be a hallucination:

  • Perfect format: Bluebook formatting is flawless
  • Real court: S.D.N.Y. (Southern District of New York) exists
  • Real reporter: F.Supp.2d is correct
  • Plausible year: 2019 is within range
  • Verification: This specific case doesn't appear in databases

CiteClear detection: May appear valid — manual verification essential

Common Hallucination Patterns

Impossible Reporter Numbers

  • F.4th, F.5th, F.6th (real: F.3d, F.4th doesn't exist as of 2024)
  • U.S. Reports > 590 (as of 2024)
  • F.Supp. > ~1000
  • F.2d > 999

Non-Existent Courts

  • 13th Circuit, 14th Circuit, etc. (only 1-11 exist)
  • Circuit Court of Appeals for [Imaginary Region]
  • D. [Non-Existent District]
  • [State] Superior Court (when state doesn't have superior courts)

Future or Anachronistic Dates

  • Cases from future years (2025, 2026, etc.)
  • Cases from years before the court existed
  • Cases from years before the statute was enacted

Generic Party Names

  • Plaintiff v. Defendant
  • Appellant v. Appellee
  • Test v. Example
  • Fake v. Imaginary
  • Corporation v. Company

Statute Extensions

  • Adding extra digits to real statute sections
  • Inventing non-existent subsections
  • Creating plausible but fake statute titles

Regulation Extensions

  • Adding .999 or similar high numbers to CFR sections
  • Inventing non-existent CFR parts
  • Creating fake agency abbreviations

The Bottom Line

Always verify AI-generated citations. Even those that look perfect can be fabricated. Use CiteClear as a first-pass filter to catch the most obvious fakes, then manually verify every citation through authoritative sources.

When in doubt, check it out. The cost of catching a fake citation before use is minimal. The cost of missing one can be your credibility, your case, or your career.

Frequently Asked Questions

Are these examples real hallucinations?

Yes, these examples are based on actual patterns observed in AI-generated legal content. While the specific citations may not have been generated by actual AI models, they follow the same hallucination patterns that AI tools commonly produce.

How can I verify if a citation is real?

Search the citation in Westlaw, Lexis, Bloomberg Law, Google Scholar, or official court databases. For statutes and regulations, use official government websites or legal research platforms. If you cannot find the citation in authoritative sources, it may be a hallucination.

Can CiteClear detect all AI hallucinations?

No. CiteClear is designed to catch the most obvious and common patterns. Some AI-generated citations may look perfectly authentic but still be fabricated. Manual verification is always essential.

What should I do if I find an AI-generated fake citation?

Remove it from your document and replace it with a verified source. Document that you found and corrected the error. Consider disclosing AI use if the document is being filed with a court or submitted to a client.

Why do AI models hallucinate citations?

AI models don't have real understanding or access to current databases. They generate text based on patterns in their training data. When asked for citations, they invent plausible-looking references that follow the patterns they've learned, without being able to verify if those references actually exist.

Are some AI models better at citations than others?

Yes, there are differences. However, all current AI models hallucinate citations to some degree. More recent models and models fine-tuned on legal data may hallucinate less frequently, but the problem persists across all major AI systems.

Can I prevent AI from hallucinating citations?

There's no guaranteed way to prevent hallucinations. You can reduce the risk by: providing more context to the AI, using more specific prompts, asking the AI to verify its citations, and using AI tools that are specifically designed for legal work. But manual verification is always necessary.

What are the consequences of using AI-hallucinated citations?

In legal contexts, consequences can include sanctions, fines, reputation damage, malpractice claims, and potential disbarment. In academic contexts, consequences can include paper retractions, loss of credibility, and disciplinary action. In business contexts, consequences can include lost clients, damaged reputation, and financial liability.

Ready to verify your own citations?

Don't let AI hallucinations slip through. Use CiteClear to catch the most obvious fakes, then verify the rest manually.

Check citations now →