Skip to content
TEN Brief Ten verified stories a day 2026.08.23 KO

이 기사는 한국어로도 읽을 수 있습니다 →

Tech · 4 min read · Reference

What an AI hallucination is — why models are trained not to say 'I don't know'

An AI hallucination is a language model producing something untrue in the form of a confident statement. Research published in 2026 locates the cause not in a defect of the model but in how models are graded: when only correctness is scored, 'I don't know' earns the same zero as a wrong answer, which makes guessing the winning strategy on anything the model does not know

A sunlit reading desk by a tall window with an open book and a neat stack of papers on pale wood

The three lines

  • Definition — an answer whose grammar and form are intact but whose content is false. The tell is that nothing in the sentence signals it is wrong
  • Cause — under accuracy-only scoring, abstaining earns the same score as a wrong answer. The model learns to guess
  • Variation — measured 2026 hallucination rates diverge sharply by task: fact-recall lowest, open-ended generation highest

Key questions

What is an AI hallucination?
A language model producing something untrue as though it were true. The point is not that it is wrong but that **nothing in the sentence signals it is wrong**. When people are unsure it usually shows — the sentence trails off, or hedges appear. A model's false answer has the same shape as a true one. Non-existent paper titles, wrong case numbers, functions that do not exist all arrive as fluent prose. That is why every article on this page carries a separate 'what could not be confirmed' section.
Why doesn't it just say it doesn't know?
Because that loses points. A 2026 paper from OpenAI researchers and a paper in Nature identify the same mechanism: most benchmarks grade **only whether the answer is correct**. Under that scheme 'I don't know' scores zero, exactly like a wrong answer, while a lucky guess scores one. So **abstaining is strictly worse than guessing** on anything the model does not know. It is the same arithmetic as an exam with no penalty for wrong answers, where nobody leaves a blank. Models are optimised against that scoring, and confident wrong answers are the result.
How often is it wrong?
It depends heavily on the task. Taken together, 2026 reporting places frontier hallucination rates broadly in the 3–19% range, while a separate benchmark across 37 models spans 15–52%. The spread comes from the task type: fact-recall questions sit low (roughly 4–13%), while open-ended generation has been measured at 40–80%. In the other direction, OpenAI's own evaluations report rates below 2% on retrieval-grounded tasks. The pattern matters more than any single number — grounding lowers it, open-endedness raises it.

Ask an AI a question and an answer comes back. The prose is smooth, something that looks like evidence is attached, and the tone is confident.

And the content is false.

This is a hallucination. Research published in 2026 traced its cause somewhere unexpected — not to the model, but to the grading sheet.

1. The definition is not "wrong"

An ordinary errorA hallucination
Sentence formbroken or truncatedintact
Confidencewaveringhigh
Detectabilityeasyhard
Examplemalformed outputa paper title that does not exist, a wrong case number, a function that was never written

The point is not that it is wrong. It is that nothing in the sentence signals it is wrong.

When people speak without confidence, it shows: the sentence trails off, hedges appear. A model's false answer has exactly the same shape as a true one. The reader has no cue to work with.

2. The cause is in the scoring

In 2026 a paper from OpenAI researchers and a paper in Nature converged on the same mechanism.

The problem is how models are evaluated. Most benchmarks grade only whether the answer is correct.

The model's answerIf correctIf incorrect
Answers what it knows1
"I don't know"00
Guesses without knowing10

Read that table from the model's position and only one conclusion follows.

Abstaining always scores zero. Guessing scores either zero or one. Therefore always guess.

It is the arithmetic of an exam with no penalty for wrong answers, where nobody leaves a blank. Models are optimised against this scoring, and the output is a confident wrong answer.

The implication is worth stating plainly: hallucination is not a model malfunctioning. It is a model doing well at what it was asked to do.

3. How often it happens

The numbers diverge sharply by task.

Measured onReported rate
Frontier models overall (2026)roughly 3–19%
A separate benchmark across 37 models15–52%
Fact-recall questionsroughly 4–13%
Open-ended generation40–80%
Retrieval-grounded tasksbelow 2%

The range is wide because benchmarks differ in their question sets and grading rules. Numbers from different benchmarks should not be placed side by side.

What to read here is not the absolute values but the pattern, and it is in the last two rows.

  • The more open the question, the higher the rate. A wide answer space is a wide space to fill in.
  • Grounding cuts it sharply. Attaching retrieved sources has been evaluated at below 2%.

4. Extended reasoning does help

Configurations that let a model think longer before answering have been reported to lower the rate.

ModelDefaultExtended reasoning
GPT-5.5 Pro8.3%4.2%
Claude Opus 4.79.4%5.1%
DeepSeek V412.7%10.4%

The third row is the important one. The improvement is not uniform. The first two nearly halved; the third fell by 2.3 percentage points. "Thinking longer halves it" holds for some cases, not as a rule.

(These figures come from secondary summaries; the underlying benchmark designs were not checked here.)

5. Why it does not simply go away

Three causes overlap.

CauseDetail
The scoring structureAs long as only accuracy is measured, guessing pays. The incentive survives until evaluation changes
Limits of training dataA fact absent from the data does not come back as "absent" but as something similar
Probabilistic generationChoosing the next token by probability makes plausibility and truth different axes

The second and third are structural and hard to change soon. The first can be changed. Add an abstention option to the grading and penalise wrong answers, and the incentive inverts. That is the direction the 2026 research proposes.

6. What a user can do

#PracticeWhy it works
1Ask for sources alongside the answergrounded tasks show markedly lower rates
2Supply the material yourselfit shrinks the blanks the model would fill
3Narrow open questionsopen-ended generation is the worst case
4Say explicitly: "if you don't know, say so"it partially inverts the scoring inside the prompt
5Verify numbers, proper nouns and quotations at sourcethese are where hallucination concentrates
6Do not treat fluency as evidence of accuracyfluency and truth are uncorrelated

Rule five is the one that catches people. Paper titles, case numbers, statistics, quotations — the items most tedious to verify are exactly the items most likely to be invented.

7. Common questions

Q. Does a bigger model fix it? It reduces it without eliminating it. No frontier model measured at zero in 2026.

Q. Can a model report its own confidence? It can emit probability values, but whether those values match actual accuracy is a separate question. This page could not confirm how well calibrated current models are.

Q. Is retrieval safe? It reduces the rate substantially but introduces a new failure. If retrieval returns the wrong document, the model builds a wrong answer on it. The presence of a citation is not itself evidence of accuracy.

Q. Is it worse in languages other than English? The general explanation is that it worsens where training material is thinner, by language or by topic. This page could not find systematic comparative measurements for Korean.

8. What is not confirmed

  • Every rate quoted here — drawn from secondary summaries; question sets, grading rules and measurement dates were not checked against the originals. Figures from different benchmarks must not be compared side by side.
  • "Extended reasoning halves it" — based on a handful of models. The third row of that table shows the limit.
  • How the Nature and OpenAI papers relate — the differences in experimental design and the extent of agreement were not checked at source.
  • Confidence calibration — how accurately current models score their own probabilities could not be confirmed.
  • Korean-language rates — no systematic measurement was found.

Sources

  1. OpenAI — Why language models hallucinate
  2. Kalai, Nachum et al. (OpenAI) — Why Language Models Hallucinate (arXiv)
  3. Nature — Evaluating large language models for accuracy incentivizes hallucinations
  4. Digital Applied — AI Hallucination Rate Benchmarks 2026: 5-Model Study
  5. SuprMind — Latest AI Hallucination Rates & Benchmarks for New AI Models, August 2026
  6. ModelsLab — LLM Hallucination Rates 2026: Best and Worst Models
  7. SQ Magazine — LLM Hallucination Statistics 2026

Verification

Published
Last modified
Cross-check
Checked against 7 independent sources.
Unverified
  • All hallucination rates given here (3–19%, 15–52%, 4–13%, 40–80%, below 2%, and the per-model extended-reasoning figures) come from secondary summaries of benchmark work. This page did not verify each benchmark's question set, grading rules or measurement date. Figures from different benchmarks should not be compared side by side
  • The claim that extended reasoning roughly halves the rate rests on a small number of models and could not be confirmed as generalising across models and tasks
  • The extent to which the Nature paper and the OpenAI paper agree, and how their experimental designs differ, could not be checked against the originals
Authoring
Reviewed by a person before publication. The full process is described in the Editorial.

Ten stories, once each morning

We send the three-line summaries only; the full pieces stay on the site. One-click unsubscribe, any time.

Related