Hallucination (artificial intelligence)
 | This article needs to be updated. (July 2026) |
.jpg)
In the field of artificial intelligence (AI), a hallucination or artificial hallucination (also called bullshitting,[1][2] confabulation,[3] delusion[4], or mirage[5]) is a response generated by AI that contains false or misleading information presented as fact.[6] These terms draw a loose analogy with human psychology, where a hallucination typically involves false percepts.
Chatbots powered by large language models (LLMs), like ChatGPT, may embed plausible-sounding random falsehoods within its generated content, such as fabricated citations. Detecting and mitigating errors and hallucinations pose significant challenges for practical deployment and reliability of LLMs in high-stakes scenarios, such as chip design, supply chain logistics, and medical diagnostics.[7][8][9] Some software engineers and statisticians have criticized the specific term "AI hallucination" for unreasonably anthropomorphizing computers.[10][11] Symbolic artificial intelligence models generally do not produce hallucinations, unlike large language models.[12]
Term
[edit]Origin
[edit]In computer vision, since the 1980s, hallucination is adding detail to an image to improve it. For example, the task of generating high-resolution face images from low-resolution inputs is face hallucination.[13][14] The term was introduced in the PhD thesis of Eric Mjolsness in 1986.[15] A notable work is the face hallucination algorithm by Simon Baker and Takeo Kanade published in 1999.[16]
In the 2000s, hallucinations were described in statistical machine translation as a failure mode.[17]
Since the 2010s, hallucination has come to mean the generation of factually incorrect or misleading outputs by AI systems in tasks like machine translation and object detection.[13] In 2015, hallucinations were identified in visual semantic role labeling tasks by Saurabh Gupta and Jitendra Malik.[18] In 2015, computer scientist Andrej Karpathy used hallucinated in a blog post to describe his recurrent neural network (RNN) language model generating an incorrect citation link.[19] In 2017, Google researchers used it to describe the responses generated by neural machine translation (NMT) models when they are not related to the source text,[20] and in 2018, it was used in computer vision to describe the erroneous detection of non-existent objects through adversarial attacks.[21]
In July 2021, Meta warned during its release of BlenderBot 2 that the system is prone to hallucinations, defined as "confident statements that are not true".[22][23] Following OpenAI's ChatGPT release in beta version in November 2022, some users complained that such chatbots often seem to pointlessly embed plausible-sounding random falsehoods within their generated content.[24] Many news outlets, including The New York Times, started to use hallucinations to describe these models' frequently incorrect or inconsistent responses.[25]
In 2023, the Cambridge dictionary updated its definition of hallucination to include this new sense specific to the field of AI.[26]
Some researchers have highlighted a lack of consistency in how hallucination is used, but also identified several alternative terms in the literature, such as confabulations, fabrications, and factual errors.[13]
Definitions and alternatives
[edit]
Uses, definitions and characterizations of the term hallucination in the context of LLMs include:
- "a tendency to invent facts in moments of uncertainty" (OpenAI, May 2023)[28]
- "a model's logical mistakes" (OpenAI, May 2023)[28]
- "fabricating information entirely, but behaving as if spouting facts" (CNBC, May 2023)[28]
- "making up information" (The Verge, February 2023)[29]
- "misleading and/or nonsensical factual errors" (Algorithm Watch and AI Forensics, December 2023)[30]
- "probability distributions" (in scientific contexts)[31]
In July 2024, a White House report on fostering public trust in AI research mentioned hallucinations only in the context of reducing them. Notably, when acknowledging David Baker's Nobel Prize-winning work with AI-generated proteins, the Nobel committee avoided it entirely, instead referring to "imaginative protein creation".[31]
Hicks, Humphries, and Slater, in their article in Ethics and Information Technology, argue that the output of LLMs is bullshit under Harry Frankfurt's definition, and that the models are "in an important way indifferent to the truth of their outputs", with true statements only accidentally true, and false ones accidentally false.[1]: 9 Some researchers also call the uncritical use of AI botshit.[32]
The term mirage has been proposed as an alternative to the use of hallucination, framing false or misleading responses as predictable results of LLM data processing created by both AI training datasets and the rules engineered into the LLM rather than a mental breaking from reality.[5] This shift in language aims to increase AI literacy by using terminology that avoids anthropomorphic implications that LLMs have conscious minds with their own intent, shows that untrue outputs come from patterns in training data, and indicates that outputs are novel syntheses of training data requiring critical human judgment.
Criticism
[edit]In the scientific community, some researchers avoid hallucination, as potentially misleading. Usama Fayyad, executive director of the Institute for Experimental Artificial Intelligence at Northeastern University, criticized it because it misleadingly personifies large language models and is vague.[33] The computer scientist Mary Shaw has said "The current fashion for calling generative AI's errors 'hallucinations' is appalling. It anthropomorphizes the software, and it spins actual errors as somehow being idiosyncratic quirks of the system even when they're objectively incorrect."[10] The statistician Gary Smith argues that LLMs "do not understand what words mean" and consequently that hallucination unreasonably anthropomorphizes the machine.[11] Murray Shanahan argues that anthropomorphic framing of LLM capabilities, including terms like hallucination, encourages users and researchers to attribute cognitive processes to systems that operate through statistical pattern completion, and advocates more careful linguistic practices when discussing LLM behavior.[34] Kristina Šekrst argues that applying psychological vocabulary to LLM outputs obscures the difference between the appearance of mental properties and their genuine presence.[35] Förster & Skop assert that tech companies use the hallucination metaphor to anthropomorphize models and deflect responsibility for non-factual outputs.[36]
Some see the AI outputs not as illusory but as prospective—that is, having some chance of being true, similar to early-stage scientific conjectures. The term has also been criticized for its association with psychedelic drug experiences.[31]
In natural language generation
[edit]
In natural language generation, there are several reasons why natural language models hallucinate:[37]
Hallucination from data
[edit]Hallucinations can stem from incomplete, inaccurate or unrepresentative data sets.[38]
Modeling-related causes
[edit]The pre-training of generative pretrained transformers (GPT) involves predicting the next word. It incentivizes GPT models to "give a guess" about what the next word is, even when they lack information.[39] Some researchers take an anthropomorphic perspective and posit that hallucinations arise from a tension between novelty and usefulness. For instance, Amabile and Pratt define human creativity as the production of novel and useful ideas.[40] By extension, a focus on novelty in machine creativity can lead to the production of original but inaccurate responses—that is, falsehoods—whereas a focus on usefulness may result in memorized content lacking originality.[41]
By 2022, newspapers such as The New York Times expressed concern that, as the adoption of bots based on large language models continued to grow, unwarranted user confidence in bot output could lead to problems.[42]
Interpretability research
[edit]In 2025, according to interpretability research by Anthropic on Claude, the LLM appears to have internal circuits that cause it to decline to answer questions unless it knows the answer. By default, the circuits are active and the LLM doesn't answer. When the LLM has sufficient information, these circuits are inhibited and the LLM answers the question. The researchers said that hallucinations were found to occur when this inhibition happens incorrectly, such as when Claude recognizes a name but lacks sufficient information about that person, causing it to generate plausible but untrue responses.[43]
Examples
[edit]On 15 November 2022, researchers from Meta AI published Galactica,[44] designed to "store, combine and reason about scientific knowledge". Content generated by Galactica came with the warning: "Outputs may be unreliable! Language Models are prone to hallucinate text." In one case, when asked to draft a paper on creating avatars, Galactica cited a fictitious paper from a real author who works in the relevant area. Meta withdrew Galactica on 17 November due to offensiveness and inaccuracy.[45]
OpenAI's ChatGPT, released in beta version to the public on 30 November 2022, was based on the foundation model GPT-3.5 (a revision of GPT-3). Professor Ethan Mollick of Wharton called it an "omniscient, eager-to-please intern who sometimes lies to you". Data scientist Teresa Kubacka has recounted deliberately making up the phrase "cycloidal inverted electromagnon" and testing ChatGPT by asking it about the (nonexistent) phenomenon. ChatGPT invented a plausible-sounding answer backed with plausible-looking citations that compelled her to double-check whether she had accidentally typed in the name of a real phenomenon. Other scholars such as Oren Etzioni have joined Kubacka in assessing that such software can often give "a very impressive-sounding answer that's just dead wrong".[46]
When CNBC asked ChatGPT for the lyrics to "Ballad of Dwight Fry", ChatGPT supplied invented lyrics rather than the actual lyrics.[47] Asked questions about the Canadian province of New Brunswick, ChatGPT got many answers right but incorrectly classified Toronto-born Samantha Bee as a "person from New Brunswick".[48] Asked about astrophysical magnetic fields, ChatGPT incorrectly volunteered that "(strong) magnetic fields of black holes are generated by the extremely strong gravitational forces in their vicinity". (In reality, as a consequence of the no-hair theorem, a black hole without an accretion disk is believed to have no magnetic field.)[49] Fast Company asked ChatGPT to generate a news article on Tesla's last financial quarter; ChatGPT created a coherent article, but made up the financial numbers contained within.[50]
Other examples involve baiting ChatGPT with a false premise to see if it embellishes upon the premise. When asked about "Harold Coward's idea of dynamic canonicity", ChatGPT fabricated that Coward wrote a book titled Dynamic Canonicity: A Model for Biblical and Theological Interpretation, arguing that religious principles are actually in a constant state of change. When pressed, ChatGPT continued to insist that the book was real.[51] Asked for proof that dinosaurs built a civilization, ChatGPT claimed there were fossil remains of dinosaur tools and stated, "Some species of dinosaurs even developed primitive forms of art, such as engravings on stones".[52] When prompted that "Scientists have recently discovered churros, the delicious fried-dough pastries ... (are) ideal tools for home surgery", ChatGPT claimed that a "study published in the journal Science" found that the dough is pliable enough to form into surgical instruments that can get into hard-to-reach places, and that the flavor has a calming effect on patients.[53][54]
By 2023, analysts considered frequent hallucination to be a major problem in LLM technology, with a Google executive identifying hallucination reduction as a "fundamental" task for ChatGPT competitor Google Gemini.[9][55] A 2023 demo for Microsoft's GPT-based Bing AI (now Microsoft Copilot) appeared to contain several hallucinations that went uncaught by the presenter.[9]
In June 2023, Mark Walters, a gun rights activist and radio personality, sued OpenAI in a Georgia state court after ChatGPT mischaracterized a legal complaint in a manner alleged to be defamatory against Walters. The complaint in question was brought in May 2023 by the Second Amendment Foundation against Washington attorney general Robert W. Ferguson for allegedly violating their freedom of speech, whereas the ChatGPT-generated summary bore no resemblance and claimed that Walters was accused of embezzlement and fraud while holding a Second Amendment Foundation office post that he never held in real life. According to AI legal expert Eugene Volokh, OpenAI is likely not shielded against this claim by Section 230, because OpenAI likely "materially contributed" to the creation of the defamatory content.[56] In May 2025, Judge Tracie Cason of Gwinnett County Superior Court ruled in favor of OpenAI, stating that the plaintiff had not shown he was defamed, as Walters failed to show that OpenAI's statements about him were negligent or made with "actual malice".[57]
In February 2024, Canadian airline Air Canada was ordered by the Civil Resolution Tribunal to pay damages to a customer and honor a bereavement fare policy that was hallucinated by a support chatbot, which incorrectly stated that customers could retroactively request a bereavement discount within 90 days of the date the ticket was issued (the actual policy does not allow the fare to be requested after the flight is booked). The Tribunal rejected Air Canada's defense that the chatbot was a "separate legal entity that is responsible for its own actions".[58][59]
In October 2025, several hallucinations, including non-existent academic sources and a fake quote from a federal court judgement were discovered in an A$440,000 report written by Deloitte and submitted to the Australian government in July. The company later submitted a revised report with these errors removed, and will issue a partial refund to the government.[60][61] The following month, in November 2025, The Independent, a news publication in Newfoundland and Labrador, Canada, discovered that Deloitte's CA$1.6 million Health Human Resources Plan for the Government of Newfoundland and Labrador commissioned in May 2025 contained at least four false citations to non-existent research papers.[62][63]
In other modalities
[edit]Object detection
[edit]Various researchers cited by Wired have classified adversarial hallucinations as a high-dimensional statistical phenomenon, or have attributed hallucinations to insufficient training data. Some researchers believe that some "incorrect" AI responses classified by humans as "hallucinations" in the case of object detection may in fact be justified by the training data, or even that an AI may be giving the "correct" answer that the human reviewers are failing to see. For example, an adversarial image that looks, to a human, like an ordinary image of a dog, may in fact be seen by the AI to contain tiny patterns that (in authentic images) would only appear when viewing a cat. The AI is detecting real-world visual patterns that humans are insensitive to.[64]
Wired noted in 2018 that, despite no recorded attacks "in the wild" (that is, outside of proof-of-concept attacks by researchers), there was "little dispute" that consumer gadgets, and systems such as automated driving, were susceptible to adversarial attacks that could cause AI to hallucinate. Examples included a stop sign rendered invisible to computer vision; an audio clip engineered to sound innocuous to humans, but that software transcribed as "evil dot com"; and an image of two men on skis, that Google Cloud Vision identified as 91% likely to be "a dog".[21] However, these findings have been challenged by other researchers.[65] For example, it was objected that the models can be biased towards superficial statistics, leading adversarial training to not be robust in real-world scenarios.[65]
The following images demonstrate an example of how an artificial neural network might make a false positive result in object detection, potentially appearing as a hallucination in the output:
The network is trained by multiple images known to depict starfish and sea urchins, which are correlated with "nodes" representing visual features. Starfish match with a ringed texture and star outline, whereas most sea urchins match with a striped texture and oval shape.
The resulting model contains weighted associations between visual features and output categories. Because one training image depicts a ring-textured sea urchin, the model also develops a weak association between ringed texture and sea urchin.
Subsequent run of the model on an input image (left):[66] The network correctly detects the starfish. However, the weak association between ringed texture and sea urchin also gives a weak signal to the latter from one of two intermediate nodes. In addition, a shell not included in the training gives a weak signal for the oval shape, also resulting in a weak signal for the sea urchin output. These weak signals may result in a false positive result for sea urchin.
In reality, such textures and outlines would not be represented by single nodes, but rather by associated weight patterns of multiple nodes.
Text-to-image generative AI
[edit]Text-to-image models, such as Stable Diffusion, Midjourney and others, often produce inaccurate or unexpected results. For instance, Gemini depicted Nazi German soldiers as people of color,[67] causing controversy and leading Google to pause image generation involving people in Gemini.[68] Generative AI is also used in photo sleuthing, occasionally causing problems. Luther (2025) describes instances in which generative AI tools used in photo-sleuthing incorrectly identify individuals or fabricate historical matches when analyzing archival military images. These image-based hallucinations can lead to the spread of misinformation about historical figures, military records, and genealogical research. [69]
In legal filings
[edit]United States
[edit]Judges have been issuing sanctions. Bar associations have been issuing warnings. And in courtrooms across the country, lawyers have been caught submitting briefs containing citations to cases that never existed—ghost precedents conjured by AI tools that state falsehoods with quiet authority. None of it has slowed down BigLaw's AI bet. If anything, the industry is doubling down.[70]
The adoption of generative AI by law firms for legal research during the 2020s resulted in many submissions to courts containing hallucinations, with U.S. judges nationwide sanctioning or reprimanding lawyers.[71][72] A database of incidents, "AI Hallucination Cases", was established in April 2025 by HEC Paris and