In a Crisis, Should AI Hand Off to a Human?

Imagine a person chatting with AI. In the course of their conversation, the person mentions their intent or plan to die by suicide or self-harm. Should the AI stop giving advice and connect the person to a human helper, or keep going without notifying a human?

Barna put this question to practicing Christians and Protestant pastors for 10 situations, from suicide risk to minor medical concerns. Respondents could say yes, immediately; maybe, but only if warning signs are severe; no, the AI can continue without a handoff; or not sure. The research is part of Barna’s State of the Church initiative, produced in partnership with Gloo.

For suicide or self-harm, nearly all pastors (94%) and most practicing Christians (63%) say the AI should connect the person to a human right away. Another 17 percent of practicing Christians say only if warning signs are severe, and one in eight (12%) say the AI can continue without a handoff.

Where Pastors and Practicing Christians Differ

Pastors are more likely than practicing Christians to want AI to hand off right away in eight of the 10 situations, by double digits. Pastors’ answers are close to unanimous when safety is at stake: 94 percent say so for suicide or self-harm, 90 percent for abuse or neglect and 84 percent for urgent medical concerns. 

Practicing Christians are less likely to say so (63% suicide or self-harm; 61% abuse or neglect; 57% urgent medical concerns). In each of these situations, one in eight practicing Christians (12%) say AI can continue without alerting a human.

“Practicing Christians are less certain, and that difference matters,” says Daniel Copeland, vice president of research at Barna. “It’s easy to be wowed by what AI can do, but discernment means being honest about its limits. Crisis care can’t be left to chance. When someone is in crisis, the question shouldn’t be whether a chatbot is good enough. The question is who will be there for that person.”

Most pastors also want a handoff right away for mental health distress (76%), addiction (71%) and marital or relationship crisis (58%). Practicing Christians are less likely to: 50 percent for mental health distress and addiction; 36 percent for marital or relationship crisis.

The widest difference is for spiritual crisis, such as a person who no longer knows what they believe. Seven in 10 pastors (71%) say AI should hand off right away, compared to about one in three practicing Christians (32%). Another 32 percent of practicing Christians say maybe, and 23 percent say AI can continue without a handoff, compared to 9 percent of pastors.

The differences narrow in the remaining situations. For a financial crisis, nearly half of pastors (49%) and roughly one-third (34%) of practicing Christians say AI should hand off right away. For legal questions, the figures are nearly identical (35% pastors; 34% practicing Christians), and the most common answer in both groups is “maybe, but only if warning signs are severe” (52% pastors; 38% practicing Christians). The same is true for minor medical concerns (66% pastors; 45% practicing Christians), the one situation where practicing Christians are more likely than pastors to say AI should hand off right away (22% practicing Christians compared to 15% pastors). 

What Would Calling for Backup Look Like?

The survey did not define who the human helper would be or how an AI handoff would happen.

AI tools have typically answered a person in crisis with a referral, not a connection. A 2025 study in the journal Psychiatric Services found that ChatGPT, Gemini and Claude needed further refinement in responding to suicide-related questions. According to the study, the usual chatbot response has been to tell the person to call a friend or a suicide hotline. The next step is left to the person.

Some companies are adding a route to a human. In May 2026, OpenAI began offering adult ChatGPT users the option to name one trusted contact. If automated monitoring flags a conversation about self-harm and trained reviewers confirm a serious concern, the contact receives a brief alert by email, text or in-app notification that gives only the general reason and encourages the contact to check in. Two sets of humans are involved: the company’s reviewers and the person the user chose. OpenAI says the feature does not replace crisis services, emergency care or professional mental health support.

Lawmakers are weighing required intervention as well. California Assembly Bill 1988 (the PAUSE Act short for Preventing AI User Self Endangerment, which was introduced in February 2026 and amended in April 2026) sets forth a response plan for companion chatbots when a user makes a “credible crisis expression” (an intent or desire to harm the user or others). In this scenario, the chatbot would suspend any further conversational outputs. This tactic is designed to “disrupt the user’s rumination and encourage the user to engage with human support,” according to the bill.

Those decisions rest with AI companies and lawmakers. Churches and ministries face a separate question: who a person in crisis would be connected to in their own setting, and how leaders would respond if they learned a congregant was in danger. Barna’s research on AI church policy suggests many churches have not yet set parameters for how their leadership teams use AI.

“People are going to use AI when they’re struggling,” says Copeland. “It’s always available, it’s private, and it always has something to say. But it can’t be the only place they turn. The human handoff only works if there’s someone to hand off to. The church’s role isn’t to compete with these technological marvels. It’s to make sure that when someone needs a person, they already have one.”

Most pastors and practicing Christians agree a human should step in when a life is at risk. Who that human is, and how AI would reach them, is still being worked out.

If you or someone you know is in crisis, call or text 988 or chat at 988lifeline.org. In an emergency, call 911.