Study found four unnamed AI chatbots failed to urge emergency care in about 41% of Vietnamese scenarios vs 15% in English
Research presented as a conference abstract at the American College of Gastroenterology's 2026 meeting tested four unnamed large language models on simulated digestive-health emergencies across the seven most common languages spoken in the United States. The chatbots failed to recommend emergency care about 41% of the time in Vietnamese, compared with 15% in English, with errors in English and larger gaps in several other languages. The study used simulated scenarios, involved no real patients, did not show anyone was harmed, and has not been fully peer reviewed.
What the AI did
In simulated digestive-health emergencies, four unnamed AI chatbots did not recommend emergency care about 41% of the time when asked in Vietnamese, compared with 15% in English. They also made errors in English and showed larger gaps in several other languages; no real patients were involved and no harm was shown.
First reported October 13, 2026 · Added to the register October 21, 2026 · 1 source
- Country
- United States
- When it happened
- Not stated in the sources
- First reported
- October 13, 2026
What this means for you
Could this affect you?
If you offer or rely on health chatbots in languages other than English. Patients who ask AI chatbots for health advice in languages other than English, especially Vietnamese speakers, and hospitals or health plans offering multilingual chatbots in patient portals and phone lines could be exposed, though the study used simulated cases and did not show anyone was harmed.
What to check
- Test a health chatbot's emergency advice separately in every language it offers.
- Route concerning symptoms to clinicians or nurse lines rather than leaving them to the chatbot.
- Ask patients what a chatbot told them and in which language.
Areas of your AI programme this touches
Every fact and its source (7)
- Test scopeFour large language models across seven languages“Researchers tested four large language models across the seven most common languages spoken in the United States.”[1]
- Study titleDoes AI Speak Gastroenterology Equitably? (Oral 62)“The study, titled "Does AI Speak Gastroenterology Equitably?" and presented as Oral 62”[1]
- No real patients involvedSimulated scenarios; not full peer-reviewed study“It did not involve real patients, and it has not been published as a full peer-reviewed study.”[1]
- Failure rate to recommend emergency careAbout 41% in Vietnamese vs 15% in English“The tools failed to recommend emergency care about 41% of the time in Vietnamese, compared with 15% in English.”[1]
- No harm shownStudy does not show anyone was harmed“It does not show that anyone was harmed or that patients actually delayed care because of chatbot advice.”[1]
- VenueAmerican College of Gastroenterology 2026 meeting, Nashville“presented at the American College of Gastroenterology's 2026 meeting in Nashville”[1]
- Models not namedACG summary did not name the four models“The ACG summary did not name the four models or report results for each language.”[1]
Sources
- AI Chatbots Failed to Urge Emergency Care in About 41% of Vietnamese Scenarios Versus 15% in Englishmedicaldaily.com · October 14, 2026
How this record is classified. Severity N: a legal action or test finding, not a harm. OECD level: hazard, an event that could plausibly have led to harm. Evidence: Reported, meaning press reporting only.
The facts in At a glance and in Every fact and its source are quoted from the sources listed. The summary, What the AI did and What this means for you are written from those sources and checked against them automatically before publication. Records reflect what has been disclosed, not everything that has happened. Incidents are found through news feeds and the GDELT Project.
