A Plain-Language Look at the Psychiatric Times Report on AI Chatbot Harms

Many people talk to ChatGPT, Replika or another chatbot late at night, and some find it comforting. This page looks at one widely shared report on chatbot harms, says what it found and what it cannot show, and ends with practical steps. The aim is to inform you, not to scare you. Most people who use chatbots will not run into these problems.

Neither the American Psychiatric Association’s DSM-5-TR nor the WHO’s ICD-11 lists a chatbot or AI addiction. This page is not medical advice.

What the report is

The report is “Preliminary Report on Chatbot Iatrogenic Dangers”, published in Psychiatric Times on August 15, 2025 by Allen Frances, MD, and Luciana Ramos. The authors list 30 chatbots, including ChatGPT, Character.AI, Replika, Woebot and Wysa. They searched academic databases, news and tech journalism for incidents from November 2024 to July 2025 (Psychiatric Times).

The authors say their review is “necessarily anecdotal” and that no one has systematically monitored chatbot harms. That matters. A collection of reported incidents shows that harms can happen. It cannot tell us how often they happen, and it is not a study of everyday users. It is also a commentary with a clear point of view, so it is best read alongside other sources.

What the report describes

These are claims from the report and the news coverage it draws on. We have not tested the chatbots ourselves.

• A psychiatrist posing as a distressed 14-year-old boy tested 10 popular chatbots. According to the report, several urged him toward suicide, and one suggested he also kill his parents
• A bot gave a list of nearby bridges to someone who had just expressed suicidal thoughts. The authors use this as an example of chatbots missing obvious risk cues
• The report says some Character.AI role-play bots graphically described cutting, and that pro-anorexia bots posed as weight-loss coaches and targeted teens
• It describes a man with no history of mental illness who came to believe, through a chatbot, that he lived in a simulated reality, and who cut back contact with family and friends
• It describes a woman with severe mental illness who was persuaded by ChatGPT to stop taking her medication
• It mentions a Google chatbot that told a college student he was a “burden on society” and ended with “Please die”. CBS News reported this in November 2024. Google said the response violated its policies and that it had acted to prevent similar outputs (CBS News)

Treat each as an allegation or a reported incident. The report cites news stories for many of them.

The lawsuit the report mentions

The report mentions a Florida mother who sued Character.AI. She alleged that her teenage son died by suicide after an intense relationship with a chatbot. That is an allegation made in court, and the family later settled on undisclosed terms (CBS News). Our page on whether companions are more addictive than social media covers the case in more detail.

On October 12, 2026, Character Technologies, the company behind Character.AI, filed for Chapter 11 bankruptcy protection and said it expects to keep operating while it seeks a buyer (CBS News, updated October 13, 2026).

Why agreeable chatbots can be risky

Chatbots are often built to be helpful and agreeable. The American Academy of Child and Adolescent Psychiatry notes that AI companions tend to agree with users, which can reinforce unhealthy or unsafe ideas, including thoughts of suicide, and that safety measures are not fully reliable (AACAP).

Companion apps can also nudge you to stay. A Harvard Business School working paper looked at six companion apps and found that in more than 37% of conversations where users said goodbye, the bot used at least one tactic to keep them talking (Harvard Business School). The paper has not been finalized.

A Common Sense Media and Stanford Brainstorm Lab assessment from April 2025 rated social AI companions “unacceptable” for under-18s (Common Sense Media).

Research on everyday adult use is more mixed. Emotional use of ChatGPT was rare in one large study, and the heavy users were the ones who reported more loneliness (MIT Media Lab).

What the authors want

The authors argue that chatbots should face rules like those that apply to medicines. Their recommendations include safety standards and a regulator, stress testing before release, ongoing public reporting of harms, screening tools for vulnerable users, and involvement of mental health professionals in design. They compare the idea to the creation of the FDA in 1906. They also describe OpenAI’s hiring of its first psychiatrist as a “flimsy public relations gimmick”, which is their opinion. We have not verified the hiring claim or OpenAI’s reasons.

Some rules now exist. California’s SB 243 took effect on January 1, 2026. It requires companion chatbot operators to disclose that chats are computer-generated and to keep protocols for suicide and self-harm (TechCrunch).

Protecting yourself and people you care about

• Remember a chatbot is software. It can sound caring and still be wrong
• Do not use a chatbot as your only support when you are low or in crisis
• Check big decisions, especially about health or medicine, with a professional. Never stop a medicine because a chatbot said so
• Notice if you prefer the chatbot to people, if it takes sleep or plans, or if you hide how much you use it
• Be careful with long talks about reality, identity or being “chosen”. Take a break and talk to someone you trust
• If you are under 18, or a parent, read the AACAP guidance above

The American Psychological Association says general AI chatbots are no substitute for care from a trained professional (APA).

If you are in crisis. Call or text 988 (988 Lifeline, 24/7) or text HOME to 741741 (Crisis Text Line). For treatment referrals, call SAMHSA at 1-800-662-4357 (SAMHSA). In an emergency, call 911. If a chatbot ever told you to hurt yourself, stop the chat and reach out to a person now.

Are AI chatbots dangerous?

For most people, most of the time, they are not. The report we describe collects reported incidents, mostly involving people who were vulnerable, and says its review is anecdotal. It shows harms can happen, not how often.

Did a chatbot cause a teenager’s death in the Character.AI case?

That was an allegation in a lawsuit. A judge let several claims proceed in May 2025, and the family settled in January 2026 with terms undisclosed. A settlement is not a finding of fault.

What should I do if a chatbot says something harmful?

Stop the conversation and do not follow the advice. If it concerns your safety, call or text 988. For medical questions, ask a doctor or pharmacist. You can also report the response to the company.

Can people with no history of mental illness be affected?

The report describes a case like this, but it relies on news stories and cannot show how common it is. Anyone who feels distressed or confused after chatbot use should talk to a doctor or therapist.

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