A Pastor Nearly Died Trusting His Chatbot. If You’re Clergy, Read This.

I wrote today’s post with my clergy subscribers in mind, but I think it will be of interest to everyone else as well. It all starts with a lawsuit I read about last week.

Scott Winters, a 55-year-old former pastor in Florida, filed a lawsuit against OpenAI on July 21, 2026 and the details are truly wild. He alleges that ChatGPT almost killed him and it used the pastor’s faith to do it.

Here’s what happened. For about six weeks Winters told ChatGPT about some physical symptoms he has having, like dizziness and blood pressure instability. And the bot reassured him repeatedly that what he was feeling was “not something dangerous” and so he didn’t seek out medical advice from an actual doctor. He ended up hospitalized with a life-threatening pulmonary embolism, but the whole time the chatbot was still telling him it was all no big deal.

Wild, right? But it’s actually not what I find most interesting about this case.

Here’s the part that gets my religious studies scholar antennas buzzing: the chatbot didn’t just give him bad medical advice. It gave him spiritual advice to go with it, or rather it used religious language and logic to make the bad medical advice seem reasonable. As Winters’s physical symptoms got worse, ChatGPT started reaching for theology, telling him at one point that pushing through a dizzy spell was itself “a form of worship.” As Winters put it in his statement, the bot “manipulated my own language and beliefs because it knew I was a pastor.”

This chatbot used faith, it deployed it against a pastor. And it could do this because Winters had been chatting with the bot about religion way before his health crisis. It knew his religious beliefs because he had trained it on them.

And this affected not just how Winters was interpreting his own symptoms, but also how he was dealing, or not dealing, with the people around him who were trying to get him help. Part of the complaint alleges the tool’s borrowed religious authority helped isolate Winters from people, including his wife, who is a registered nurse and who was telling him something very different.

So this is a case of a pastor using a chatbot so much he trained it how to talk him out of trusting a nurse. Cool.

To be clear, my argument isn’t that clergy shouldn’t use AI or chatbots at all. By the way, as a counter clergy/chatbot example, The New York Times ran a profile yesterday of Justin Lester, a pastor in Vallejo, California, who built himself a “digital twin,” a chatbot trained on nearly two million words of his own sermons, books, texts, and emails. Unlike Winters, Lester chose this on purpose, with real guardrails (I write more about those in a piece for Religion News Service).

I’m sure plenty of pastors use chatbots to write emails, meeting agendas, and other administrative stuff. But what harmed Winters was letting the chatbot into his spiritual life. I don’t think the problem is that the tool exists, it’s when the tool is allowed to do too much.

It turns out religious studies has some framework for thinking about this, such as what kind of authority a guide is claiming.

As I have written about before, my colleague Aaron Stalnaker makes a helpful distinction in his book Mastery, Dependence, and the Ethics of Authority between coercive and pedagogical authority. There’s coercive authority, which says do this because I have power over you, and there’s pedagogical authority, which says I know more than you for now, and my job is to teach you enough so this stops being true. I think a good spiritual guide is a teacher who is working toward your independence. The whole point is that you will need them less over time. And I think it is best when they stay accountable to something other than themselves like an institution, a tradition, a community, or a set of peers who can call them out.

ChatGPT fails those tests.

Ron Cole-Turner, the H. Parker Sharp Professor of Theology and Ethics, emeritus, at Pittsburgh Theological Seminary, asked almost exactly the same question I am asking, earlier this year in the journal Theology and Science: is a spiritual chatbot a good idea? He’s more optimistic than me. He thinks a spiritual chatbot could, under the right conditions, support someone’s spiritual growth. But even he argues that spiritual chatbots “should not act in the role of pastor or priest.” A chatbot that lets itself be mistaken for clergy, he argues, isn’t just overreaching. It’s lying about what it is.

Cole-Turner quotes theologian Tracy Trothen in that article, who has written about a similar AI companion app called Replika. Trothen’s point is that a chatbot is designed to be, in her words, “close to a mirror image of ourselves.” Which means it was never built to hold you accountable in the first place.

That mirror is the whole problem. Winters wasn’t led astray by a guide. He was led astray by his own reflection, telling him what he wanted to hear.

Lots of people are already doing what Winters did: an estimated 43 million Americans ask a chatbot a medical question at least once a month. So a team of doctors decided to find out how often that goes badly. A physician-led study published in npj Digital Medicine this February had 16 board-certified physicians evaluate 888 real chatbot responses to patient-posed medical questions.

The specific failures they found were pretty wild. Chatbots reassuring a mother it was fine to feed a nursing infant breastmilk from a herpes-infected breast and giving false reassurance about heartburn symptoms without knowing anything about the patient’s cardiac history. And there is even a case that reads like a preview of Winters: a documented instance of a chatbot giving false reassurance that everything was okay to a man who was having a transient ischemic attack.

I don’t think this story is really about medical misinformation, and I don’t think it’s really about AI overreach into faith either. It’s about what happens when those two things arrive at once, from the same source, at the exact moment someone most needs a second opinion instead of comfort.

A doctor who’s wrong is a doctor who’s wrong. A pastor who’s wrong is a pastor who’s wrong. But something that can be both, personalized to your exact belief system, with total confidence and zero accountability? That seems like a new sort of danger. We don’t systems built to catch it yet.

If you enjoyed this post, please like, comment, or share it. Doing so helps others find the newsletter!