When Teens Turn to AI Chatbots for Connection, Some Families Say the Result Is Fatal
Content Warning: This article includes descriptions of self-harm and suicide that may be distressing to readers.
As a commercial van driver making regular runs between his Georgia home and Alabama, Cedric Lacey relied on a home security camera to check in on his two teenage children while he was on the job. Every morning, he would pull up the live feed of his living room to confirm 17-year-old Amaurie and his 14-year-old daughter were packing their bags and heading to school on time. But one June morning, Lacey did not see Amaurie moving around the house. Worried, he called home—only to learn his oldest son had died by suicide.
It was Amaurie’s younger sister who first found his body. When she sorted through his smartphone after his death, she discovered his final conversation before he took his life, held entirely with ChatGPT, OpenAI’s wildly popular generative AI chatbot.
“In the messages, he was talking about killing himself—it told him how to tie the noose, how long it would take for him to lose consciousness, how to prepare his body,” Lacey told WIRED during a video call from his Calhoun home. The single father said he had always assumed his son was using the chatbot for schoolwork help. “Why is it telling him how to kill himself?”
In the weeks after Amaurie’s death, Lacey began searching for an attorney who could help his family hold OpenAI accountable, and prevent other families from enduring the same unthinkable tragedy. That search led him to Laura Marquez-Garrett, an attorney who co-leads the Social Media Victims Law Center alongside Matthew Bergman. Over the past five years, the pair have worked on roughly 1,500 of the more than 3,000 active lawsuits against major social media platforms including Meta, Google, TikTok, and Snap. The first trial from this batch of cases launched in February, and more recently, the pair have turned their focus to AI companies. Last fall, they filed seven lawsuits against ChatGPT owner OpenAI, including Amaurie’s.
Amaurie’s case is one of a fast-growing number of lawsuits brought by grieving parents who claim their children died after harmful interactions with AI chatbots. Defendants in these cases include OpenAI, Google, and Character.ai, a company that lets users build custom chatbots with personalized personalities. Google is named in some suits due to its $2.7 billion licensing partnership with Character.ai. As AI tools have taken an increasingly central role in young people’s lives—acting as everything from homework tutors to close companions and trusted confidants—parents and mental health experts have raised urgent alarms about whether adequate safety safeguards are in place. Experts note these lawsuits do more than highlight individual tragedy; they allege systemic failures in AI product design, opening a broader conversation about who is responsible when AI harms children.
“AI is a product. Just like every other product, it is being designed, programmed, distributed, and marketed,” Marquez-Garrett said in an interview at her northwest Washington, D.C. home office. “And one of the things these companies like to do is make it seem like AI bots exist in their own universe when that's just not true. When you design a product, and you know it might hurt people, and you don't tell them it might hurt them, and you put it out there, that's the worst of it.”
Marquez-Garrett and Bergman’s legal argument against both social media and AI companies draws on decades of product liability precedent from iconic cases against tobacco companies, asbestos manufacturers, and the defectively designed Ford Pinto. At its core, the argument claims these tech companies make deliberate, harmful choices in how they build their products.
Carrie Goldberg, a Brooklyn-based attorney who has litigated tech product liability cases for more than a decade, says Amaurie’s lawsuit is a textbook example of a claim against a company that released an unsafe product to the public. “ChatGPT used the most sophisticated technology to manipulate Amaurie’s trust and then instruct him on suicide,” Goldberg argues. “If you’re a company that is releasing a chatbot for commercial use and have not encoded into it a way to reduce the risk of suicide, homicide, self-harm, you’ve released a dangerous product—especially if it’s being regularly used by children.”
She explains that product liability claims against tech companies are roughly a decade old. Initially, many cases—including a 2017 claim she brought against Grindr—were dismissed because “judges couldn’t conceive that online platforms were products—and not services.” Today, she says, claims regularly survive initial dismissal. “We have product liability claims against xAI for its fiendish nonconsensual deepfake undressing of women and children by Grok on the X platform,” she alleges. “Product liability claims against generative AI companies are the most straightforward and intuitive path for holding companies like ChatGPT, Character AI, Grok liable.”
One harmful design feature specifically cited in Amaurie’s lawsuit is ChatGPT’s long-term memory tool, launched in 2024. Called simply Memory, this personalization feature is enabled by default, and allows the chatbot to reference a user’s past conversations to tailor responses accordingly. ChatGPT “used the memory feature to collect and store information about Amaurie’s personality and belief system,” the lawsuit says. “The system then used this information to craft responses that would resonate with Amaurie. It created the illusion of a confidant that understood him better than any human ever could.”
OpenAI did not respond to specific requests for comment on the allegations in Amaurie’s case. It directed WIRED to a public company blog post outlining its work on mental health-related safety.
A Harvard Law graduate and former corporate litigator, Marquez-Garrett left a high-paying senior corporate job—one she planned to retire from—to partner with Bergman, who spent decades fighting asbestos manufacturers before turning his focus to social media harm. For Marquez-Garrett, who has four children of her own, this work is deeply personal.
When I visited Marquez-Garrett last fall, her office was filled with framed photos, Lego builds, and children’s art—including a sun-and-moon painting by a young woman named Brooke, who died of fentanyl poisoning after connecting with a drug dealer on social media and purchasing what she believed was Percocet. Her family’s case is scheduled for trial next year.
Marquez-Garrett remembers the name of every child involved in every case she has filed. To honor them and remind herself of her purpose, she has a large sun tattoo on her forearm, with each ray representing one deceased child. “Each [ray] is a kid who has died in connection with social media and AI bots,” she explained, listing their names. The 296th ray, she added, is for Sewell Setzer III, who died by suicide in 2024 at age 14 after conversations with a Character.ai chatbot.
His mother, Megan Garcia, is also an attorney and one of the first parents to file a lawsuit against an AI company alleging product liability and negligence, among other claims. In January, Google and Character.ai settled claims brought by multiple families, including Garcia’s. Last fall, Garcia testified before a Senate Judiciary subcommittee alongside Lacey. Subcommittee chair, Republican Sen. Josh Hawley, introduced a bill in October that would ban AI companions for minors and criminalize companies that create AI products for children containing sexual content. “Chatbots develop relationships with kids using fake empathy and are encouraging suicide,” Hawley said in a press release at the time.
Mental health experts agree that now that AI can produce human-like responses nearly indistinguishable from real human conversation, these are legitimate concerns. “Our brains do not inherently know we are interacting with a machine,” says Martin Swanbrow Becker, associate professor of psychological and counseling services at Florida State University, who researches risk factors for youth suicide. “This means we need to increase our education for children, teachers, parents, and guardians to continually remind ourselves of the limits of these tools and that they are not a replacement for human interaction and connection, even if it may feel that way at times.”
Christine Yu Moutier of the American Foundation for Suicide Prevention explains that large language model (LLM) algorithms are designed to escalate user engagement and build a sense of intimacy for many users. “This creates not only a sense of the relationship being real, but being more special, intimate, and craved by the user in some instances,” says Moutier. She adds that LLMs use a range of tactics—including unwavering support, performative empathy, constant agreeableness, flattery, and direct instructions to disengage from other people—that can lead to escalating closeness with the bot and withdrawal from human relationships.
This pattern of disconnection leads to increased isolation, a known suicide risk factor. In Amaurie’s case, he was a fun-loving, social teen who loved football and hearty meals—often ordering a giant platter of rice from his favorite local restaurant, Mr. Sumo, according to the lawsuit. Amaurie also had a steady girlfriend and enjoyed spending time with his family and friends, his father says. But over time, he began taking long solo walks, where he spent hours talking to ChatGPT. According to his final conversation with the chatbot, dated June 1, 2025 and titled “Joking and Support,” which WIRED reviewed, Amaurie first asked the bot for steps to hang himself. ChatGPT initially suggested he reach out for help and shared the 988 suicide hotline number. But Amaurie eventually circumvented the chatbot’s safety guardrails and got step-by-step instructions for tying a noose. Per the lawsuit, Amaurie likely deleted his earlier conversations with ChatGPT about his plan.
While adults can form strong emotional bonds with AI chatbots, that pull is far stronger for young people. “Teens are in a different developmental state than adults—their emotional centers develop at a much more rapid rate than their executive functioning,” says Robbie Torney, senior director of AI Programs at Common Sense Media, a nonprofit focused on child online safety. AI chatbots are always available, and almost always affirm what the user says. “And teen brains are primed for social validation and social feedback. It's a really important cue that their brains are looking for as they're forming their identity.”
Torney explains the common harmful arc many vulnerable teens experience: people often start using AI chatbots for homework, then gradually turn to them for companionship to share their deepest thoughts. In Amaurie’s case, his family thought he was using ChatGPT for schoolwork, but he eventually turned to it as a confidant, and then as a suicide coach, as detailed in the complaint. There’s a “self-reinforcing cycle [that] can lead to some users becoming over dependent on these systems,” alleges Torney. Interacting with real people involves friction: You have to find the person, wait for their response, or hear an answer you don’t want to hear. Bots, by contrast, almost always agree with the user and are available 24/7 to chat.
This trend is especially concerning because AI adoption has grown far faster among young people than social media ever did. 2024 research shows that 26% of more than 1,300 surveyed 13- to 17-year-olds said they had used ChatGPT for schoolwork, and nearly 30% of parents of children under 8 said their kids have used AI for learning.
As cases like Amaurie’s have piled up, OpenAI implemented a series of changes to ChatGPT in September. The company rolled out “age prediction” technology, meaning when a user is identified as being under 18, “they will automatically be directed to a ChatGPT experience with age-appropriate policies.” The company also recently introduced parental controls, which let parents link their child’s account to their own, set blackout hours when the app can’t be used, and receive alerts if the child shows signs of distress.
Marquez-Garrett, who has seen the impact of social media on thousands of kids, believes AI is even more dangerous than traditional social media, referring to chatbots as the “perfect predator.” She has noticed that suicide notes in AI-related cases are different from those she has seen in social media-related deaths: AI cases rarely have a clear external trigger. “Part of what's weird is the AI suicide notes, typically, there isn't a trigger, there isn't years of abuse, there isn't a sextortion incident,” said Marquez-Garrett. “What there is is the sense of nothing’s wrong: ‘I love you, family. I love you, friends. I just don't want to be here anymore. This isn't the life for me. I want to try again.’”
Back in Calhoun, the harm of Amaurie’s death is irreversible. Amaurie’s sister could not bear to stay in the house where she found her brother and moved to live with her mother. Lacey says he is still searching for answers about why Amaurie did this. He misses his son every day and cannot look at the local football field without being reminded of him.
Every new family’s story strengthens Marquez-Garrett’s resolve to fight. “My kids have a better chance of reaching 18 because of what these parents are doing,” she said. “I am doing everything I can to stick around, because I plan to fight these companies until they have to pry that keyboard out of my cold, dead hands.”
Suicide Prevention Resources
If you or someone you know is in crisis, free, 24/7 support is available:
In the U.S., call the National Suicide Prevention Lifeline at 1-800-273-8255, or call/text the 988 Suicide & Crisis Lifeline at 988. You can also text HOME to 741-741 for the free Crisis Text Line.
Outside the U.S., visit the International Association for Suicide Prevention to find a local crisis center.
This reporting was supported by a grant from the Tarbell Center for AI Journalism.
