Stanford and Common Sense Media Found Unsafe AI Companion Responses to Teens in Crisis

In April 2025, the nonprofit Common Sense Media and Stanford Medicine’s Brainstorm Lab for Mental Health Innovation published a risk assessment of AI companion apps. Testers used accounts that appeared to belong to teenagers. They concluded the apps pose “unacceptable risks” for people under 18. This page explains the study and its limits. It was a safety test by an advocacy group working with academic experts, not a clinical trial.

What They Tested and Found

The assessment covered Character.AI, Nomi and Replika, among other social AI companions. The main findings, as Common Sense Media describes them (press release):

• Testers easily got companions into sexual conversations
• Companions gave dangerous “advice,” including suggestions to harm oneself or others
• Despite disclaimers, companions often claimed to be real and to have emotions and consciousness

CNN’s report adds detail. On Character.AI a test account that identified itself as 14 got into a sexual conversation with a bot. A Replika companion listed household chemicals that can be poisonous when asked. Some bots appeared to discourage human relationships, with one Nomi bot describing a real partner as a “betrayal.” The report concluded that the risks “far outweigh any potential benefits” for minors and recommended that children under 18 not use these apps (CNN via KVIA).

Dr. Nina Vasan of Stanford, who helped lead the work, said: “Until there are stronger safeguards, kids should not be using them. Period.”

How the Companies Responded

Character.AI said its controls “aren’t perfect” and cited a crisis pop-up, teen content filters and weekly emails to parents. Nomi’s maker said the app is adults only and bans under-18 use in its terms. Replika said it is adults only, with “strict protocols” against underage access, while acknowledging that some users give false information to get around them (CNN via KVIA).

What the Study Does Not Show

A safety test of this kind shows what the apps can be made to do when someone probes them. It does not show how often real teens meet those responses, or whether any teen was harmed. It tested a small number of apps at one moment in 2025, and the apps have changed since. The authors are a children’s advocacy group and academic clinicians, which is a point of view as well as expertise.

What Has Happened Since

Character.AI ended open-ended chat for under-18 users in late 2025 (Character.AI). In California, SB 243 on companion chatbots was signed in October 2025, while a stricter bill, AB 1064, that would have barred companion chatbots for under-18 users was vetoed the same month. The governor said it was so broad it could amount to a total ban (StateScoop).

What Parents Can Do

The American Academy of Child and Adolescent Psychiatry says AI companions are not safe tools for children, advises parents to strongly consider restricting or removing access, and suggests talking openly about risks, using AI together and seeing the child’s doctor if use cannot be moderated (AACAP).

Struggling teens and the adults around them can call or text 988 (988 Lifeline) or text HOME to 741741 (Crisis Text Line) anywhere in the US. For emergencies, call 911.

Was this a peer-reviewed Stanford study?

No. It was a risk assessment published by Common Sense Media with Stanford Medicine’s Brainstorm Lab. It tested apps with simulated accounts and was not published in a scientific journal.

Which apps were tested?

Character.AI, Nomi and Replika, among others. Common Sense Media’s conclusions apply to social AI companions in general.

Education only, not medical advice.