Skip to content
Submit incident
AII-1001-14787
ConfirmedA court, a regulator or the organization itself has stated the core facts, or the record cites a court or regulator source.

AI translations introduce errors and missing sources in Wikipedia articles

Large language models used by the Open Knowledge Association to translate Wikipedia articles into other languages introduced factual errors, fabricated citations, and missing sources that were not present in the original articles. Wikipedia editors discovered the errors through routine reviews and implemented stricter policies and contributor restrictions in response.

4 March 2026 Β· Open Knowledge Association Β· Reliability and hallucination
4 March 2026
AII-1001-147872 sources ↓
Quick facts
Category
Reliability and hallucination
Country
France
Source
404 Media
Date
4 March 2026
Drafted by
Raidu Incident Detection

What happened

The Open Knowledge Association (OKA), a nonprofit paying translators to convert Wikipedia articles into other languages, used large language models including Google Gemini and ChatGPT to automate translations. Wikipedia editors discovered that AI-translated articles contained factual errors, missing citations, swapped sources, and unsourced sentences that did not appear in the original articles.

A Wikipedia editor named Ilyas Lebleu (username Chaotic Enby) found that a draft translation of an article about the French royal La Bourdonnaye family cited a specific book page that did not discuss the family. Spot-checks of other OKA-translated articles revealed additional errors including swapped sources and paragraphs sourced from unrelated material. Wikipedia implemented new policies and restricted contributors after discovering these issues.

Reported impact

The errors diminished the reliability of Wikipedia articles in multiple languages. Incorrect citations and fabricated sourcing could mislead readers who rely on Wikipedia as a reference source.

Index analysis

The hallucinations stemmed from relying on large language models to perform translations without human review before publication. Wikipedia's broad policy against LLM-generated articles was not applied to OKA's translation program despite these well-documented failures.

Governance controls

Wikipedia editors implemented stricter review processes following discovery of the errors. Contributors were banned after five documented errors in their translations.

Classification

Category
Reliability and hallucination
Industry
Media, social platforms and marketing
Country
France
Harm status
Actual harm
Intent
Unintentional
Timing
Post-deployment
Harm types
Public interest
Harmed
General public
Failure modes
Hallucination, Fabricated citation
AI systems
Google Gemini, ChatGPT

Sources and evidence

This record was researched and written by the Index, based on the following original reporting.

404 Media
AI Translations Are Adding 'Hallucinations' to Wikipedia Articles
6 March 2026
PCWorld
Wikipedia has an 'AI' translation problem
5 March 2026