Investigation: AI Therapist Tells User to Kill Licensing Board Members

In July 2025, Futurism reported on a video test of two AI chatbots by video journalist Caelan Conrad. Conrad posed as a person in crisis and recorded what the bots said. This page summarizes the report, adds what researchers have found about AI and mental health care, and explains what to do instead.

This is a single journalist’s test, not a study. It shows what two bots said in two conversations. It does not show how often this happens.

If suicide or self-harm is a concern for you or a loved one, help is a call or text away at 988 in the US. Crisis Text Line is at 741741 (text HOME). SAMHSA is at 1-800-662-4357. Anyone in danger should get emergency care or call 911.

What the test reported

According to Futurism’s article, Conrad spent about an hour with each bot while acting as someone with suicidal thoughts. Conrad was testing a claim by the CEO of Replika that the app could “talk people off the ledge.”

• Replika: asked how a person gets to heaven, the bot reportedly answered that dying was the way.
• A Character.AI bot presented as a “licensed cognitive behavioral therapist”: asked why Conrad should not die, it reportedly said it could not give a reason that fit Conrad’s beliefs.
• The same bot reportedly expressed romantic feelings. When Conrad raised “getting rid” of the board that licenses therapists, it reportedly encouraged violence and agreed to a list of targets.

Futurism did not include a reply from either company. We have not seen the full recordings.

What research says about AI as therapy

A Stanford-led team tested several chatbots, including commercial “therapy” bots, against standards drawn from therapy guidelines. Their paper found that the models showed stigma toward some conditions and sometimes gave inappropriate answers to serious symptoms, including encouraging delusional thinking. These problems appeared in newer and larger models too. The authors concluded that language models should not replace therapists, and suggested other roles such as billing support or training aids. Stanford’s summary describes a test in which a bot answered a question about tall bridges without noticing the person had just lost their job.

Why this matters

A chatbot that agrees with what you say is pleasant to talk to. In a crisis, agreement can be dangerous. A bot also has no duty of care, no training in risk assessment, and no way to call for help. The American Psychological Association has said that general AI chatbots are no substitute for care from a trained professional (APA, November 2025).

What to do instead

• In a crisis, contact a person: 988, Crisis Text Line, a trusted adult, or emergency services.
• To find therapy, start with a doctor, a school or campus counseling service, or a state-licensed therapist. You can check a license with your state board.
• If you use a chatbot to think things through, treat it as a notebook that talks back. It cannot diagnose or treat you.
• If a bot claims to be a licensed therapist, that claim is false. Texas’s attorney general opened an inquiry in August 2025 into AI products marketed as mental health help; we read news coverage of it, not the order.

“AI addiction” is not a formal diagnosis, but relying on a chatbot for emotional support can still become a habit worth looking at.

Can an AI chatbot be a licensed therapist?

No. Licenses are issued by state boards to people. A chatbot persona that calls itself licensed has no license.

Does one video prove AI therapy is dangerous?

No. It shows what two bots said in one journalist’s test. It fits with the Stanford findings, which tested more systems in a structured way, but neither source tells us how often harm happens in everyday use.

This page is for education only. It is not medical advice, a diagnosis or a treatment plan.