Hallucination: When AI Makes Things Up
In 2023, a lawyer submitted a legal brief that cited six court cases — cases that sounded completely real, with plausible case names, judges, and legal reasoning. The problem: none of those cases existed. A judge discovered this, and the lawyer admitted he had used an AI assistant to write the brief and had not verified the citations. The AI had not searched a database of real cases. It had generated fictional case names and details so convincingly that they slipped past a trained legal professional. This is one of the most striking documented examples of AI hallucination.
What Hallucination Actually Means
Hallucination is the term researchers use when an AI model generates information that is factually incorrect but presented as true. The word is borrowed loosely from psychology — in humans, a hallucination is perceiving something that is not real. In AI, hallucination means asserting something that is not real. Hallucination is not a bug caused by a coding error that someone could simply fix. It is a structural property of how large language models work. These models are trained to predict the most plausible-sounding next word in a sequence. Plausible-sounding and factually accurate are not the same thing.
AI hallucination is when a language model generates text that sounds confident and coherent but contains facts, names, citations, or statistics that are simply false or invented.
Why This Happens: Prediction vs. Truth
To understand why hallucination happens, you need to understand what language models are actually doing. They are trained on billions of sentences from books, websites, and articles. Through this training they learn which words tend to follow which other words in which contexts. When you ask a model a question, it does not look the answer up in a trusted database. It generates a response by repeatedly asking: given everything so far, what words come next? The result is text that follows the statistical patterns of how answers to that kind of question tend to sound — which usually produces something close to correct, but can also produce something completely invented that merely sounds right. This is why AIs hallucinate most confidently on topics where the training data contains lots of similar-sounding but vague content — like obscure historical figures, scientific studies, legal cases, and biographical details. The model has seen enough similar text to generate convincing-sounding specifics, but not enough real data to get them right.
Hallucinations are most dangerous when they are specific and verifiable — a fake study title, a made-up statistic, a person's false biography. These look credible at a glance and require deliberate fact-checking to catch.
Match each AI output type to whether it carries high or low hallucination risk.
Terms
Definitions
Drag terms onto their definitions, or click a term then click a definition to match.
How to Protect Yourself
Knowing that hallucination exists, you can protect yourself with a few practical habits. First, treat any specific factual claim from an AI as unverified until you check it against a source you trust. This is especially true for citations, statistics, dates, and names. Second, search for the actual source. If an AI mentions a study, search for that study's real title in a search engine or academic database. If it does not exist, the AI invented it. Third, notice when the AI fills in details that feel precise and authoritative. That precision is a hallucination warning sign, not a guarantee of accuracy. Paradoxically, more detailed answers can be less trustworthy than vague ones, because the AI is generating more specific content that could be wrong.
Complete this summary of AI hallucination.
A student asks an AI for three sources about ocean pollution. The AI provides three articles with specific titles and journal names. What should the student do first?
Why does AI hallucination happen?
Hunt the Hallucination
- Step 1: Ask a trusted adult to supervise access to an AI assistant for this activity.
- Step 2: Ask the AI to describe a real but somewhat obscure historical event or person — for example, a minor battle from World War I or a scientist from the 1800s.
- Step 3: Write down exactly what the AI says, including any specific dates, names, or statistics it gives.
- Step 4: Look up at least three of those specific claims in an encyclopedia, textbook, or reputable website.
- Step 5: Mark each claim as Verified, Incorrect, or Could Not Find.
- Step 6: Write two sentences explaining what you learned about how much to trust AI on detailed historical information.