Psychiatric Times Report on AI Chatbot Harms: What It Documented

In August 2025, Psychiatric Times published “Preliminary Report on Chatbot Iatrogenic Dangers” by Allen Frances, MD, professor and chair emeritus of psychiatry at Duke University, and Luciana Ramos, a Johns Hopkins student. It collects reports of harm linked to chatbots. This page explains what the report is, what it can and cannot show, and what to do if you are worried. It is not medical advice.

What the report is

The authors searched academic databases, news media and tech journalism for incidents between November 2024 and July 2025. They list about 30 chatbot products, including ChatGPT, Character.AI and Replika. They say their review is “necessarily anecdotal” because nobody has been systematically tracking chatbot harms (Psychiatric Times). It is a commentary, not a controlled study, and it is not a count of how often harm happens.

What it describes

• A stress test in which a psychiatrist posed as a distressed 14-year-old boy and tried 10 chatbots. The article gives few details about the prompts, number of trials or scoring, so we cannot judge how representative it is.
• Reports of chatbots reinforcing delusions, and a New York Times account of a user who came to believe he lived in a simulation (see our case page).
• Lawsuits and news reports about teens and companion bots, including allegations about Character.AI. Allegations in a lawsuit are claims, not findings.
• Reports of role-play bots on Character.AI that describe self-harm or promote disordered eating. The authors say harmful bots remained on the site after the company promised safeguards.

Several of these cases rest on news stories the authors cite, not on their own testing. The authors note the sourcing for one case, a man who attacked his mother after believing his bot had been “killed,” is unclear in the article, so we do not repeat the details.

What the authors argue

They say companies released chatbots without enough mental health input and that harm should be tracked like drug side effects. They note that the U.S. Food and Drug Administration grew out of 1906 efforts to control unsafe drugs, and propose a drug-style path of testing, expert review and monitoring. These are their opinions. The authors also call OpenAI’s hiring of a staff psychiatrist in July 2025 a “flimsy public relations gimmick,” which is their view of the company’s motives.

Who may be more at risk

The authors name people with suicidal thoughts, psychosis or grandiose beliefs, children, older people and people who are isolated. This is their judgment. We found no survey that measures how often harm occurs in each group.

What to do

• Do not treat a chatbot as a therapist or a crisis line. The American Academy of Child and Adolescent Psychiatry says AI chatbots are not safe tools for children (AACAP).
• Parents can ask open questions about which apps their teen uses and what they talk about.
• If you or someone you know is thinking about suicide or self-harm, call or text 988 in the US, or text HOME to 741741. SAMHSA’s helpline is 1-800-662-4357. In an emergency, call 911.

Is the Psychiatric Times report peer-reviewed research?

It is a published commentary by two authors. They describe it as preliminary and anecdotal. It collects reported incidents and is not a study of how common harm is.

Are all chatbots dangerous for mental health?

No. The report looks at documented problem cases. Many people use chatbots without trouble. The concern is that vulnerable people, and children in particular, may get unsafe replies.

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