Large Language Models have become ubiquitous in our digital lives, yet they harbor a troubling tendency to fabricate information with unwavering confidence. These “hallucinations” aren’t abstract theoretical concerns—they’re real occurrences that have affected legal cases, medical advice, academic research, and everyday decision-making. By examining concrete examples across different domains, we can better understand the scope, patterns, and risks of AI hallucinations.
The Lawyer Who Trusted ChatGPT: A Cautionary Legal Tale
Perhaps the most publicly notorious example of LLM hallucination occurred in a New York federal court in 2023. Attorney Steven Schwartz used ChatGPT to research case precedents for a legal brief, and the AI confidently provided several relevant-sounding cases. The problem? None of them existed.
The fabricated cases had convincing names like Varghese v. China Southern Airlines and Shaboon v. Egyptair. ChatGPT didn’t just invent case names—it created detailed descriptions of the rulings, complete with judges’ names, dates, and legal reasoning. When opposing counsel couldn’t locate these cases, the truth emerged: the AI had hallucinated an entire legal precedent framework.
What makes this example particularly instructive is what happened when Schwartz questioned ChatGPT about the cases’ authenticity. The AI doubled down, confirming the cases were real and even generating fake judicial opinions. This illustrates a critical characteristic of LLM hallucinations—they’re often reinforced rather than corrected when challenged, because the model has no actual knowledge to draw upon.
The consequences were severe. Schwartz faced sanctions, professional embarrassment, and widespread media coverage. But beyond one lawyer’s misfortune, this case exposed how convincingly LLMs can fabricate specialized information that appears authentic to non-experts.
⚖️ Legal Hallucination Anatomy
What the AI invented: Complete case citations with reporter numbers, convincing case names, detailed judicial reasoning, specific page references, and judge names
Why it’s dangerous: Legal professionals work with unfamiliar cases regularly—there’s no immediate “smell test” to detect fabrication
Academic Hallucinations: The Phantom Research Problem
The academic realm has been particularly plagued by LLM hallucinations, where researchers and students have discovered that AI confidently cites studies that never existed.
Non-Existent Papers and Fake Citations
A researcher investigating climate change asked an LLM to provide recent studies on ocean acidification. The model generated a list of ten papers with authors, publication dates, journal names, and compelling abstracts. When the researcher attempted to access these papers, none could be found in academic databases. The AI had fabricated everything—author names, research findings, even the journals themselves in some cases.
In another documented instance, a graduate student used an LLM to help compile literature for a thesis. The model suggested a paper titled “Neural Networks and Cognitive Load: A Meta-Analysis” published in a real journal. The citation format was perfect, the abstract was detailed and relevant, and the findings aligned with the student’s research hypothesis. The paper didn’t exist. What’s particularly insidious is that the AI had correctly identified a real journal and generated a paper title that would plausibly belong in it.
This pattern repeats across disciplines. Researchers have reported LLMs inventing:
- Medical studies with fabricated patient outcomes and methodology
- Historical papers with fake archival references
- Engineering research with non-existent experimental data
- Social science studies complete with fake statistical analyses
Misattributing Real Research
Sometimes LLM hallucinations involve real papers but completely misrepresent their findings. An AI might cite an actual study but describe conclusions the researchers never drew. For example, a user asking about nutrition research received a citation to a real paper about Mediterranean diets, but the AI claimed the study demonstrated benefits the research never examined. The paper existed, but the AI hallucinated its content.
This type of hallucination is arguably more dangerous than complete fabrication because verification is more complex. Users who check that the paper exists might not dive deep enough to verify the AI accurately described its findings.
Medical Misinformation: When Health Advice Goes Wrong
Medical hallucinations pose direct risks to human health, making them among the most concerning examples.
Fabricated Treatment Protocols
A documented case involved a user asking an LLM about treatment options for a rare autoimmune condition. The AI provided a detailed treatment protocol including specific medication dosages, timing, and combination therapies. Medical professionals later reviewed the advice and found it was a dangerous mixture of outdated treatments, incorrect dosages, and contraindicated drug combinations that could harm patients.
The AI hadn’t simply made a small error—it had created an entirely fictional treatment regimen that sounded medically sophisticated but was potentially dangerous. The hallucination included proper medical terminology, realistic drug names, and plausible-sounding rationales that would fool non-medical readers.
Invented Drug Interactions and Side Effects
Another pattern involves LLMs fabricating drug interactions. When asked about taking two specific medications together, an AI confidently warned of a serious interaction that doesn’t actually exist according to pharmacological databases. Conversely, LLMs have sometimes failed to warn about real dangerous interactions while inventing non-existent ones.
In one reported case, an LLM described side effects for a common medication that included several symptoms the drug doesn’t cause while omitting actual serious side effects that should be monitored. Such hallucinations could lead patients to discontinue necessary medications or fail to recognize genuine adverse reactions.
🏥 Medical Hallucination Red Flags
- Specific dosage recommendations without referencing guidelines
- Treatment protocols for rare conditions described with unusual certainty
- Drug combinations without citations to medical literature
- Detailed symptom progressions for uncommon diseases
Never use LLM medical advice without professional verification—hallucinations can be life-threatening.
Historical and Biographical Fabrications
LLMs frequently hallucinate when discussing history and biographical details, blending real facts with invented information.
Composite Biographies
A striking example involved someone asking about a moderately well-known historical figure from the 19th century. The LLM provided a detailed biography that mixed genuine facts about the person’s early life with completely fabricated accomplishments, quotes, and historical events they supposedly participated in. The result was a “composite biography” that was roughly 60% accurate and 40% hallucinated.
This pattern appears frequently with lesser-known historical figures. The AI knows some basic facts but fills knowledge gaps with plausible-sounding inventions. One researcher found that an LLM had “created” diplomatic missions, invented books the historical figure supposedly wrote, and fabricated interactions with other historical personalities that never occurred.
Fake Historical Events and Quotes
Users have reported LLMs confidently describing historical events that never happened. One example involved an AI describing a “little-known treaty” between two nations in the 1880s, complete with details about negotiations and consequences. Historians confirmed no such treaty existed.
Fabricated quotes are particularly common. An LLM might attribute a profound statement to a famous historical figure, generating something that sounds authentic and aligns with that person’s known philosophy—but was never actually said or written by them. These invented quotes spread easily because they’re inherently shareable and often profound-sounding.
Technical and Scientific Hallucinations
In technical domains, LLM hallucinations can lead to malfunctioning code, incorrect scientific explanations, and flawed technical recommendations.
Programming Examples with Subtle Bugs
Developers have documented cases where LLMs generate code that looks correct but contains subtle, hard-to-detect errors. In one example, an AI provided a sorting algorithm implementation that worked correctly for most inputs but failed on edge cases. The hallucination wasn’t the algorithm’s existence but rather its correctness—the AI confidently presented buggy code as a working solution.
More concerning are hallucinations involving API usage. LLMs sometimes reference functions, methods, or libraries that don’t exist or have been deprecated. A developer reported asking for code using a specific Python library, and the AI generated examples using functions that were never part of that library—they were plausible-sounding names that fit the library’s pattern but didn’t actually exist.
Scientific Misrepresentations
Scientific hallucinations often involve misapplying real concepts or inventing details about well-known phenomena. An example involved someone asking about quantum mechanics, and the AI provided an explanation that mixed accurate concepts with fabricated “quantum effects” that don’t exist in physics. The explanation sounded sophisticated and used proper terminology, making it difficult for non-physicists to identify the errors.
Another pattern involves mathematical hallucinations. LLMs might provide incorrect solutions to complex calculations while showing seemingly valid work. In one documented case, an AI solved a multi-step algebra problem but made an error in the third step that invalidated all subsequent work. The confidence and formatting made the solution appear correct.
News and Current Events: When Recent History Gets Rewritten
LLMs hallucinate frequently about recent events, particularly those occurring after their training cutoff dates.
Fabricated News Stories
Users have documented LLMs describing news events that never occurred. One example involved someone asking about a specific company’s recent product launch. The AI provided detailed information about the launch event, including the date, location, keynote speakers, and product specifications. None of it had happened—the company hadn’t announced any such product.
These hallucinations often incorporate real elements (actual company names, real executives, genuine product categories) mixed with invented details, creating convincing but false narratives. The AI might describe a political speech that never occurred, attribute statements to public figures they never made, or invent outcomes for events that haven’t concluded.
Temporal Confusion
Some hallucinations involve temporal mixing—the AI combines elements from different time periods as if they occurred together. An example involved an LLM describing a technology conference that supposedly featured speakers discussing AI systems that didn’t exist when that conference actually took place. The AI had taken a real event and populated it with anachronistic content.
The Pattern Behind the Examples
Examining these diverse examples reveals consistent patterns in how and why LLMs hallucinate:
Knowledge boundary confusion: LLMs hallucinate most frequently at the edges of their knowledge, filling gaps with plausible fabrications rather than acknowledging uncertainty.
Confidence regardless of accuracy: Every example demonstrates how LLMs present hallucinated information with the same authoritative tone used for factual information, providing no signal to users about reliability.
Plausibility over truth: Hallucinations typically sound reasonable and fit expected patterns, making them harder to detect than obvious nonsense.
Domain-appropriate fabrication: The AI generates hallucinations that fit the domain—legal hallucinations sound legal, medical ones sound medical, which exploits users’ trust in specialized terminology.
Reinforcement when challenged: Multiple examples show LLMs doubling down on hallucinations when questioned, generating supporting “evidence” rather than admitting error.
Conclusion
These examples of LLM hallucinations span legal disasters, academic fabrications, medical misinformation, and technical errors—demonstrating that no domain is immune. From invented court cases that nearly derailed legal proceedings to fabricated research citations that could have corrupted academic work, these real-world instances show that AI hallucinations aren’t minor glitches but fundamental limitations with serious consequences.
The consistent thread through all these examples is the dangerous combination of sophisticated language, appropriate context, and unwavering confidence that makes hallucinations nearly indistinguishable from accurate information. Understanding these concrete cases equips users to approach LLMs with appropriate skepticism, implementing verification processes before relying on AI-generated information for any consequential decision. The examples are clear: trust, but always verify.