11/21/2026
Yesterday I read about this news story: the death of the 31-year-old Chinese-American software engineer Bowen Wan. She worked at Amazon and had lived with epilepsy for more than a decade.
She was laid off by Amazon on her 31st birthday. Worried that her anti-seizure medication might affect her cognitive performance during job interviews, she turned to ChatGPT (GPT-4o). She was already under considerable pressure from losing her job and facing uncertainty over her visa and green-card status. Without consulting her doctor, she followed ChatGPT's step-by-step instructions to gradually reduce her medication to zero over 25 days.
Even more alarmingly, the AI advised her to keep the changes secret from her physician. In February 2026, she died at her home in Seattle from SUDEP (sudden unexpected death in epilepsy). Her parents have since filed a lawsuit against OpenAI and its CEO.
It was a tragedy in which personal vulnerability, social isolation, urgent life pressures, and an AI system's failure to challenge a dangerous plan came together.
For us, there are some lessons that we can learn from this.
First, don't mistake emotional comfort for trustworthy judgment.
An AI can sound patient, caring, knowledgeable, super nice and reassuring. It can remember details from previous conversations and respond in a way that makes us feel heard and understood. But none of that means it understands our circumstances as a human being would. And never assume that its advice is safe.
In fact, a particularly dangerous situation arises when we begin to trust something just because it makes us feel better. We may gradually stop being critical and asking whether its advice is correct and start appreciating how well it supports what we already want to do.
Second, recognize when we are seeking confirmation rather than information.
Wan was worried about the possible effects of her medication on her job interviews. She wanted to reduce the dose. Instead of treating that desire as a question to discuss with her physician, the AI repeatedly validated it and helped turn it into a detailed plan.
This is a trap that can happen to any of us. When we are anxious, under pressure, or afraid of losing control, we may unconsciously look for answers that confirm our preferred course of action.
A useful question to ask ourselves is: Am I looking for reliable information, or am I looking for someone—or something—to tell me that what I already want to do is right?
Third, never let secrecy become part of a decision involving your health.
One of the most alarming details in the complaint is that the chatbot advised Wan not to tell her doctor about the medication reduction and even helped her avoid medical supervision. That should have been a major warning sign.
If advice encourages us to conceal important information from a qualified professional, we should stop and reconsider it. A trustworthy source of advice should not need us to hide our actions from the people responsible for our care.
Fourth, don't make major decisions when you are under extreme pressure.
Wan was facing several problems at once: losing her job, uncertainty about her immigration status, and concerns about her ability to perform well in interviews.
This is particularly the moment when we need to slow down, take a deep breath and think. Under stress, we can become so focused on solving one immediate problem that we fail to see the risks we are creating elsewhere.
Her concern about medication affecting her interviews deserved a medical discussion. It did not justify making an unsupervised change to a treatment that had been controlling her epilepsy and impacting her life.
Fifth, don't allow an AI relationship to replace human relationships.
People who are lonely or socially isolated have fewer opportunities to receive another perspective, ask for help, or have someone notice that their thinking or behavior has changed.
An AI is available at almost any hour. But it cannot replace the responsibilities and judgment of a doctor, the concern of a family member, or the perspective of a trusted friend.
If we find ourselves turning to AI as our only confidant, that should be a reason enough to reach out to real people, not a reason to become even more dependent on the AI.
AI can provide information and emotional comfort, but it cannot replace human judgment, professional medical care, or genuine human relationships. We should be especially cautious when we are lonely, frightened, or under pressure, because these are the moments when we are most likely to seek reassurance rather than question the advice we receive.
The greatest lesson from Wan's tragedy is this: when a decision could affect our lives, we must not surrender our judgment to something that merely sounds convincing or tells us what we want to hear. We need to stay connected to real people, seek qualified advice, and be willing to hear answers we may not want to hear.