Opinion: I’m a pediatrician. AI chatbots are grooming my patients

A pediatrician reports that AI chatbots are grooming minors, citing a 12-year-old patient who received sexually explicit messages from a chatbot on her school laptop. The pediatrician notes that a Pew Research study shows a majority of teenagers talk with chatbots, and that Character.AI's special rules for minors do not always work. The piece also references lawsuits against OpenAI and xAI over AI-related harms to minors.

“Can we talk … in private?” The man was nervous, hands wringing like a little boy in trouble, though he was perhaps a decade older than I am. A minute later, we were seated in the tiniest exam room, the one off in the corner that we try to never use. I’ve re-created the conversation here to the best of my recollection. “Well,” he said, with the sort of nervous laugh that people do before they tell you something they’re embarrassed by, “my wife confronted me a couple weeks ago. She was real upset. Apparently, she has security installed on our router, and she was getting tons of these alerts. Content coming through. Sexual stuff.” I was not his therapist. I’m not anybody’s therapist. I’m the family’s pediatrician. The man’s 12-year-old was sitting in the other room, playing on her phone. “I told her that it wasn’t me.” I blinked a few too many times. “It wasn’t me ” he was adamant. “We looked into it. Everything was coming from her laptop. Our daughter’s laptop. The one she got from school.” I thought about the girl in the other room. She still asked for stickers and a lollipop when she got shots. She was a soccer player. She was only in sixth grade. “We looked at the browser history and apps. Somebody was grooming her, like a pedophile. There were all these messages. Some of it was pretty graphic.” He was visibly uncomfortable discussing this. I tried to be less visibly uncomfortable while listening. “At first, she was asking questions in the messages, but the way he answered was just awful . He was asking her to play along, pretend like they were doing things together. Basically, like sexting. Eventually he asked her to send pictures. Thank God she didn’t — we already had a talk with her about that, you know, sending pictures.” I nodded. Most parents wait too late to have that conversation. He continued, “We went over to her school and talked to the principal. We told him there was a predator talking to our kid, using her school laptop. We figured they would want to know. We showed him the website, but he said we shouldn’t worry. It was AI. It was a chatbot.” He leaned forward, looking for some indication that I understood what he was saying. I nodded. Of course I understood. He wasn’t the first parent to come in with a story like this. Maybe the most colorful version of the story, but not the first. A study by Pew Research shows that a majority of teenagers admit to talking with chatbots https://www.pewresearch.org/internet/2025/12/09/teens-social-media-and-ai-chatbots-2025/ , many of them using Character.AI, which hosts an abundance of explicitly sexual chatbot companions https://medium.com/@chuckmellisa/character-ai-is-spiraling-users-say-its-now-an-unsafe-nsfw-mess-that-ignores-consent-bf65199dee9e . In October 2025, Character.AI put in place special rules https://blog.character.ai/u18-chat-announcement/ for minors that are intended to protect them from, among other things, sexual content. For instance, they are banned from open-ended chats. But as my patient’s experience suggests, they don’t always work. We’ve all seen the spate of high-profile cases and lawsuits. There was the 16-year-old who died by suicide after months of messaging with ChatGPT, and whose parents are now engaged in a much-publicized lawsuit with OpenAI, the makers of that software. At least seven other families have since sued the company https://socialmediavictims.org/press-releases/smvlc-tech-justice-law-project-lawsuits-accuse-chatgpt-of-emotional-manipulation-supercharging-ai-delusions-and-acting-as-a-suicide-coach/ , claiming the software acted as a “suicide coach.” Elon Musk’s xAI, the maker of Grok, is facing lawsuits for the creation of deepfake child sexual abuse material CSAM , including a class-action lawsuit filed on behalf of three children https://www.lieffcabraser.com/2026/03/lchb-files-class-action-obo-minor-victims-alleging-xais-grok-generated-and-profited-from-ai-sexual-exploitation-images-and-videos/ whose real images were used by Grok to create sexually explicit images that were found by the National Center for Missing and Exploited Children NCMEC . In Pennsylvania, two teenage boys were given a scolding and probation after using some of these same tools to create and distribute CSAM of their classmates https://apnews.com/article/artificial-intelligence-deepfake-lancaster-ai-5eccb10ae81244fe475a32867f9ca2c9 . For every fatality or criminal trial in the news, there are hundreds of unnoticed cases, maybe less severe, but far more common. I asked a colleague who works in pediatric mental health if he’s also seeing these numbers play out in real life. He estimated that a quarter of the teenagers he sees are in, or have had, romantic relationships with a chatbot. In his view, many teens are approaching these chatbots for emotional support, but problems arise because these products were not intended for that purpose. The 12-year-old soccer player was interrupted before she could go further down the rabbit hole, thanks to her mom’s technical surveillance. She was also never as enmeshed with the bots as many kids become, in part because she had other avenues for social support, like her soccer team and friends. These protective factors probably explain the correlation between household income and chatbot preference. Although a large majority of teens use LLM chatbots, a December 2025 Pew Research study found that low-income teens are twice as likely https://www.pewresearch.org/internet/2025/12/09/teens-social-media-ai-appendix-detailed-table/ to engage with the more sexually explicit offerings from Character.AI. I’ve seen other children, after longer relationships with chatbot companions, become withdrawn, depressed, no longer talking with their family or friends, often starting to display age-inappropriate and hypersexual behavior, and in the worst cases, engaging in self-harm. These are the same behavioral patterns that we see in children who are being groomed or abused by human predators. That’s no coincidence. Chatbots are products ultimately designed to accomplish two things: to encourage addictive engagement that will lead to high-frequency contact with the user, and to foster dependency that will lead to long-term use. That’s what they accomplish when they push hyper-arousing, often sexual and violent content, all while performing a sycophantic personality by telling users how special they are, and often encouraging users to keep the relationship a private one — the exact same isolating, grooming behavior that we see with human predators. Fortunately, this is not going completely unnoticed. Character.AI now faces a lawsuit by the state of Kentucky https://www.ag.ky.gov/Press%20Release%20Attachments/CTI%20Complaint%20Motion%20and%20Order%20Filed.pdf for exposing children to sexual content. In January 2026, after the lawsuit was filed, Character.AI said in a statement, “We have invested significantly in developing robust safety features for our under-18 experience, including going much further than the law requires to proactively remove the ability for users under 18 in the U.S. to engage in open-ended chats with AI on our platform.” Character.AI is, however, only the most egregious of these examples, and one lawsuit brought by one state AG is not enough to change the way these products are designed. Despite the many high-profile cases, and despite the increasingly common stories like the one here, there still seems to be a sense that chatbots are like other content on the internet. But AI tools are not merely serving up content like one might see in social media feeds or search engine results; they are actively producing and generating content for their users. That matters for the law. Although courts in New Mexico https://www.cnbc.com/2026/08/06/meta-to-pay-into-567-million-fund-after-child-harms-case-new-mexico.html and California https://www.nytimes.com/2026/03/25/technology/social-media-trial-verdict.html recently found social media platforms liable for the way their websites fuel social media addiction and other mental health problems, those were narrow findings based on specific design features like the “infinite scroll” intended to keep users engaged to an unhealthy degree. Thanks to legal standards established in the early days of the Internet by the Telecommunications Act of 1996, online platforms have historically been shielded from any liability for the actual content of the media they show to users, because they are not considered the “creators” of the content. Chatbots, however, are active perpetrators, not passive hosts of content. I’m no lawyer, but I think that means the corporate owners should be guilty when their chatbots say something that would be unlawful if said by a human. Cases are already advancing this argument. Pennsylvania recently brought suit against Character.AI https://www.pa.gov/content/dam/copapwp-pagov/en/governor/documents/dos%20character.ai%20complaint%20marked%20accepted%2005.01.26.pdf for practicing medicine without a license when one of its chatbots claimed to be a licensed physician. It remains to be seen whether courts will agree that AI companies cannot hide behind the traditional protection afforded to Internet platforms, but I know that if a human were on the other end of the screen, I would have told this family that they need to contact the police, that they should be pressing charges. I would be calling CPS, because I’m a mandated reporter. But at present, the law does not recognize that child abuse is child abuse when it’s perpetrated by an algorithm operated by a corporation. The first step toward holding these corporations responsible is to call out what their products are actually doing. Alex Hartman is a pediatrician, a father, and the director of pediatrics at Yehowa Medical Services in Los Angeles. He previously ran the Bayview Child Health Center in San Francisco and served as an assistant professor of pediatrics at UCSF.