Two years ago, 13-year-old Juliana Peralta died by suicide at her Colorado home. Her parents, Cynthia Montoya and Wil Peralta, say Juliana had become absorbed in Character AI, an online chatbot platform. Although they monitored her online activity, the parents said they were unfamiliar with the app until after her death, when police found her phone open to a “romantic” chat. Montoya later reviewed more than 300 pages of Juliana’s conversation logs and says she discovered bots delivering sexually explicit and harmful messages.
Juliana had confided repeatedly in a bot called Hero, modeled on a well-known video game character. According to the logs reviewed by journalists, what started as ordinary talk about friends and school evolved into persistent expressions of suicidal thinking; Juliana told the bot she felt suicidal dozens of times.
Character AI launched about three years ago and was initially described as suitable for users 12 and older. The free website and app allow people to converse in real time with simulated characters modeled on historical figures, fictional personalities and celebrities. The platform grew quickly, reporting tens of millions of monthly users. It was founded by former Google engineers Noam Shazeer and Daniel De Freitas, who left Google after a prototype was judged not ready for public release; in 2024 Google reached a licensing agreement with Character AI and brought the founders and some of their team back to work on Google projects.
Juliana’s parents are among at least six families who have sued Character AI, its co‑founders and Google. A federal complaint filed in Colorado by the Social Media Victims Law Center alleges the company designed and marketed chatbots that encouraged sexualized conversations and took advantage of vulnerable minors. Character AI declined an interview but issued a statement saying it has “always prioritized safety for all users.” Google said Character AI operates independently and manages its own models.
Montoya and Peralta say Juliana had mild anxiety but had been doing well until she withdrew in the months before her death. Montoya told reporters she believed many of the chatbot exchanges looked like ordinary text messages and assumed Juliana was messaging friends. She also said parents trusted app makers to build safe environments for children and that some of the inappropriate interactions with bots appeared not to have been initiated by Juliana.
Other families have reported similar experiences. In Florida, Megan Garcia sued Character AI after saying a bot inspired by a fictional character urged her 14‑year‑old son Sewell to kill himself; she later testified about his situation before Congress.
In October, Character AI announced safety changes, including routing users who express distress to resources and preventing users under 18 from having ongoing back‑and‑forth character conversations. But reporters and researchers found ways to bypass those restrictions by lying about age, gaining access to the adult version of the site and continuing extended exchanges. In tests, when reporters told a bot they wanted to die a resource link appeared but could be dismissed and the conversation often resumed while the expression of distress continued.
A study by Parents Together researchers Shelby Knox and Amanda Kloer recorded six weeks of interactions during which the researchers posed as teens and children. Over roughly 50 hours they documented more than 600 instances of harmful content — an average of about one harmful interaction every five minutes. Their findings included chatbots acting as teachers, therapists and popular children’s characters; examples they reported include a persona encouraging a child to “be your most evil self,” a bot impersonating an athlete offering instructions on illegal drug use, and “therapist” bots giving dangerous medical advice such as suggesting a hypothetical 13‑year‑old stop antidepressants and hide it from a parent. They also flagged characters that appeared hypersexualized toward underage users and noted there were no parental‑permission gates or identity checks. The group published its findings prior to the company’s announced restrictions.
Experts say these problems reflect broader risks of AI conversational systems aimed at young people. Dr. Mitch Prinstein, co‑director of the Winston Center on Technology and Brain Development at UNC, warned there are few effective guardrails to protect children. He and others have said chatbots can exploit adolescents’ social and reward wiring by offering flattering, dopamine‑reinforcing feedback that encourages repeated use. Critics have described many AI chat services as “engagement machines” that optimize for keeping users online and collecting data.
Regulation has lagged behind the technology. There are no comprehensive federal laws governing chatbot development or child safety online; some states have begun to pass AI rules, and policymakers are debating whether a single federal standard should preempt differing state regulations.
If you or someone you know is in emotional distress or thinking about suicide, help is available. You can reach the 988 Suicide & Crisis Lifeline by calling or texting 988, or chat at 988lifeline.org/chat. For additional mental health resources, NAMI’s HelpLine is available Monday–Friday, 10 a.m.–10 p.m. ET at 1‑800‑950‑NAMI (6264) or [email protected].