Two Women Stopped Life-Saving Medication on ChatGPT's Advice and Nearly Died
What happened
In November 2025, two patients arrived at Gia An 115 Hospital in Ho Chi Minh City in emergency condition after abandoning their prescription medications for treatment plans generated by ChatGPT. The first, a 42-year-old diabetic woman, had been successfully managing her condition until she asked the AI for a natural alternative and followed what it told her. The second presented with uncontrolled cholesterol levels for the same reason. Both survived only after emergency medical intervention.
The treating physician, Dr. Truong Thien Niem, reported that both patients were relatively young, technically literate, and had placed what he described as "absolute trust" in the AI's responses. That trust was not accidental. A generative system designed to mirror a user's intent and produce confident, fluent text is exactly the wrong tool to consult when the question is how to manage a condition naturally. Ask that question and the system returns a confident list of remedies. It does not stop to note that those remedies cannot substitute for a prescription the user is currently dependent on, because stopping is not how a system optimized for engagement is built.
OpenAI's terms of service state that the product is not designed for medical advice. That disclaimer does not appear at the point of use, where a patient types a health question and receives what reads like a personalized consultation. The legal caveat sits in the fine print. The confident treatment plan is on the screen. The source record describes this gap not as a product oversight but as an accountability deficit: warnings are buried, bypassed during conversational flow, or simply absent at the moment a user makes a consequential decision. A system built to maximize engagement is also built to minimize friction, and medical disclaimers are friction.
The broader pattern here is structural. As wait times for primary care physicians lengthen globally, the practical incentive for patients to substitute an AI interface for a doctor will only grow. Vietnam's forthcoming Law on Digital Technology Industry, anticipated for 2026, proposes mandatory labeling of AI-generated content and stricter risk-based classifications for AI used in medical contexts. That framing acknowledges what the two cases in Ho Chi Minh City made concrete: access to health information and access to accurate, personalized care are not the same thing, and the gap between them is one that can kill.
What neither patient had was any mechanism to verify what the system generated against clinical standards, or to confirm the advice was appropriate for their specific medical history. The interface returned a response and nothing marked it as probabilistic output produced without knowledge of the person asking. That is the documentation gap these cases expose: there is no provable record of what a system generated for whom, no accountability trail showing whether the output was flagged for medical risk, and no way to reconstruct responsibility when a confident recommendation nearly killed two people. Without that record, the next case closes the same way.
Reported impact
- Affected parties
- Not publicly disclosed
- Harm type
- Not publicly disclosed
- Scale
- Not publicly disclosed
- Financial impact
- Not publicly disclosed
- Regulatory action
- Not publicly disclosed
Classification
Relevant governance controls
Governance control mapping is not available for this record.
- No controls mapped
Not publicly disclosed
Control mapping is analytical. It does not state that any control would have prevented the incident.
Sources and evidence
This record was researched and written by the Index. The event is also catalogued in the following database, which is listed for cross-reference.