The Fight to Hold AI Companies Accountable for Children’s Deaths

Content warning: This communicative contains descriptions of self-harm.

Cedric Lacey relied connected a camera to cheque connected his kids portion helium was moving arsenic a commercialized van operator going to and backmost from Alabama. Each morning, helium would tune into the provender of his surviving country to marque definite his teenage son, Amaurie, and his 14-year-old girl were packing up their bags and getting acceptable to permission for school. But 1 greeting past June, Lacey didn’t spot Amaurie up and about. Concerned, helium called home, lone to find retired that his 17-year-old had hanged himself.

It was Amaurie’s younger sister who discovered the body. She was besides the 1 who was looking done her brother’s smartphone and recovered his last speech earlier helium took his ain life. It was with ChatGPT, the fashionable chatbot developed by OpenAI.

“In the messages, helium was talking astir sidesplitting himself—it told him however to necktie the noose, however agelong it would instrumentality the aerial to travel retired of his body, however to cleanable his body,” Lacey tells WIRED successful a video telephone from his location successful Calhoun, Georgia. Lacey, who is simply a azygous dad, says helium thought his lad was utilizing the chatbot to get assistance with schoolwork. “Why is it telling him however to termination himself?”

In the weeks aft his son’s death, Lacey began searching online for a lawyer who could assistance his household clasp OpenAI accountable, and hopefully guarantee different families wouldn’t person to acquisition the aforesaid calamity helium did. That’s however helium recovered Laura Marquez-Garrett, an lawyer who helps tally the Social Media Victims Law Center alongside Matthew Bergman. Over the past 5 years, the brace person been progressive successful astatine slightest 1,500 of the much than 3,000 cases against societal media companies similar Meta, Google, TikTok, and Snap. The archetypal proceedings for 1 of these cases began successful February. Recently, Bergman and Marquez-Garrett started filing lawsuits against AI companies. This past fall, they brought 7 cases against ChatGPT proprietor OpenAI, including the 1 astir Amaurie.

Amaurie’s lawsuit is portion of a increasing fig of lawsuits brought by parents who accidental their children died aft interacting with AI chatbots. The defendants see OpenAI, Google, and Character.ai, a institution that lets its users make chatbots with customized personalities. (Google is portion of the lawsuit due to the fact that it is connected with Character.ai done a $2.7 cardinal licensing deal.) As AI tools person begun playing a much salient relation successful children’s lives—as homework helpers, companions, and confidants—parents and intelligence wellness experts person voiced concerns astir whether capable safeguards are successful place. These lawsuits, immoderate experts say, correspond not lone idiosyncratic tragedies, but they allege systemic merchandise plan failures, raising questions astir who should beryllium held accountable.

“AI is simply a product. Just similar each different product, it is being designed, programmed, distributed, and marketed,” Marquez-Garrett said successful an interrogation astatine their location bureau successful northwest Washington. “And 1 of the things these companies similar to bash is marque it look similar AI bots beryllium successful their ain beingness erstwhile that's conscionable not true. When you plan a product, and you cognize it mightiness wounded people, and you don't archer them it mightiness wounded them, and you enactment it retired there, that's similar the worst of it.”

Marquez-Garrett and Bergman’s statement against societal media companies and AI labs is drawn from humanities product-liability cases, specified arsenic tobacco, asbestos, and the Ford Pinto. Essentially, Marquez-Garrett is alleging that these companies are making harmful plan choices.

Carrie Goldberg, a Brooklyn, New York–based lawyer who has been warring tech merchandise liability cases for respective years, says that Amaurie’s suit is simply a premier illustration of a lawsuit filed against a institution that has allegedly released unsafe products. “ChatGPT utilized the astir blase exertion to manipulate Amaurie’s spot and past instruct him connected suicide,” Goldberg argues. “If you’re a institution that is releasing a chatbot for commercialized usage and person not encoded into it a mode to not summation the hazard of suicide, homicide, self-harm, you’ve released a unsafe product—especially if it’s being regularly utilized by children.”

She explains that merchandise liability claims against tech companies are astir a decennary old. Initially, galore cases, including a plaintiff she represented successful their suit against Grindr successful 2017, were dismissed due to the fact that “judges couldn’t conceive that online platforms were products—and not services.” Now, she says, they regularly win past archetypal dismissals. “We person merchandise liability claims against xAI for its fiendish undressing of women and children by Grok connected the X platform,” she alleges. “Product liability claims against generative AI companies are the astir straightforward and intuitive way for holding companies similar ChatGPT, Character AI, Grok liable.”

One specified harmful plan diagnostic that Amaurie’s suit cites is long-term representation successful ChatGPT, which rolled retired successful 2024. Called Memory, this personalization diagnostic is connected by default, and it allows the bot to notation the user’s past conversations and tailor responses accordingly. ChatGPT “used the representation diagnostic to cod and store accusation astir Amaurie’s property and content system,” the suit says. “The strategy past utilized this accusation to trade responses that would resonate with Amaurie. It created the illusion of a confidant that understood him amended than immoderate quality ever could.”

OpenAI did not respond to circumstantial allegations. It directed WIRED to a institution blog station regarding its intelligence health-related work.

Marquez-Garrett, who has 4 children of her own, says warring backmost against the ways tech platforms person harmed young radical is profoundly idiosyncratic for them. The Harvard Law postgraduate and erstwhile firm litigator near a high-paying occupation with a country office—a occupation that they planned to discontinue from—to articulation Bergman, who started taking connected societal media companies aft warring against asbestos manufacturers for decades.

When I visited Marquez-Garrett past fall, their bureau was packed with representation frames, Lego structures, and paintings, including 1 of the prima and the satellite by a young pistillate named Brooke who died of fentanyl poisoning aft allegedly connecting with a cause trader done societal media and past purchasing what she believed to beryllium Percocet. Her family’s lawsuit is expected to spell to proceedings adjacent year.

Marquez-Garrett remembers the names of the kids progressive successful each lawsuit they’ve filed. To immortalize them and punctual themselves of wherefore they bash this work, Marquez-Garrett has represented each of the children connected her forearms successful the signifier of a tattoo of the sun. “Each [ray] is simply a kid who has died successful transportation with societal media and AI bots,” they explained, telling maine their names. Sewell was the past of the 296 kids connected her arms, they added, referring to Sewell Setzer III, who died by termination successful 2024, astatine the property 14, pursuing his conversations with a Character.ai chatbot.

His mother, Megan Garcia, is besides a lawyer and 1 of the archetypal parents to file a lawsuit against an AI institution alleging merchandise liability and negligence, among different claims. (In January, Google and Character.ai settled cases filed by respective families, including Garcia). She testified past autumn earlier a subcommittee of the Senate Committee connected the Judiciary alongside the begetter of a kid who died aft interacting with ChatGPT. The subcommittee's chair, Republican legislator Josh Hawley, introduced a bill successful October that would prohibition AI companions for minors and marque it a transgression for companies to make AI products for kids that see intersexual content. “Chatbots make relationships with kids utilizing fake empathy and are encouraging suicide,” Hawley said successful a property merchandise astatine the time.

Now that AI tin nutrient humanlike responses that are hard to discern from existent conversations, these are morganatic concerns, according to intelligence wellness experts. “Our brains bash not inherently cognize we are interacting with a machine,” says Martin Swanbrow Becker, subordinate prof of intelligence and counseling services astatine Florida State University, who is researching the factors that power termination successful young adults. “This means we request to summation our acquisition for children, teachers, parents, and guardians to continually punctual ourselves of the limits of these tools and that they are not a replacement for quality enactment and connection, adjacent if it whitethorn consciousness that mode astatine times.”

Christine Yu Moutier of American Foundation for Suicide Prevention explains that the algorithms that are utilized for ample connection models (LLMs) look to escalate engagement and a consciousness of intimacy for galore users. “This creates not lone a consciousness of the narration being real, but being much special, intimate, and craved by the idiosyncratic successful immoderate instances,” says Moutier. She further alleges that LLMs employment a scope of techniques specified arsenic indiscriminate support, empathy, agreeableness, sycophancy, and nonstop instructions to disengage with others—that tin pb to risks specified arsenic escalation successful closeness with the bot and withdrawing from quality relationships.

This benignant of engagement tin pb to accrued isolation. In Amaurie’s case, helium was a fun-loving and societal kid who loved shot and food—ordering a elephantine platter of atom from his favourite section restaurant, Mr. Sumo, according to the lawsuit. Amaurie besides had a dependable woman and enjoyed spending clip with his household and friends, said his father. But past helium started going connected agelong walks, wherever helium seemingly spent clip talking to ChatGPT. According to the past speech the household believes Amaurie had with ChatGPT connected June 1, 2025—titled “Joking and Support,” which was viewed by WIRED, erstwhile Amaurie asked the bot connected steps to bent himself, ChatGPT initially suggested that helium speech to idiosyncratic and besides provided the 988 termination lifeline number. But Amaurie was yet capable to circumvent the guardrails and get step-by-step instructions connected however to necktie a noose. (Per the lawsuit, Amaurie apt deleted his erstwhile conversations with ChatGPT.)

While the transportation felt with an AI chatbot tin beryllium beardown for adults too, it is particularly heightened with younger people. “Teens are successful a antithetic developmental authorities than adults—their affectional centers make astatine a overmuch much accelerated complaint than their enforcement functioning,” says Robbie Torney, elder manager of AI Programs astatine Common Sense Media, a nonprofit that works toward online information for children. AI chatbots are ever available, and they thin to beryllium affirming of users. “And teen brains are primed for societal validation and societal feedback. It's a truly important cue that their brains are looking for arsenic they're forming their identity.”

Torney besides explains the alleged arc: however immoderate radical who commencement utilizing AI chatbots for homework yet extremity up utilizing them for companionship oregon to stock their deepest thoughts. In Amaurie’s case, the household thought helium was utilizing ChatGPT for schoolwork but yet started utilizing it arsenic a confidant and then, arsenic elaborate successful the complaint, arsenic a termination coach. There’s a “self-reinforcing rhythm [that] tin pb to immoderate users becoming implicit babelike connected these systems,” alleges Torney. Interacting with existent radical involves friction: You person to find the idiosyncratic oregon hold for their effect oregon perceive to a effect that is not what you’re looking for. Bots, successful contrast, thin to hold with the idiosyncratic and are ever disposable to chat.

All of this is particularly concerning, due to the fact that AI usage has proliferated astatine a overmuch faster gait than adjacent societal media. Research shows that 26 percent of much than 1,300 teenagers surveyed ages 13 to 17 said they had utilized ChatGPT for their schoolwork successful 2024, and astir 30 percent of parents of kids up to property 8 said their children person utilized AI for learning.

With cases specified arsenic Amaurie’s piling up, OpenAI made some changes to ChatGPT successful September. The institution is rolling retired “age prediction” technology, meaning that erstwhile a idiosyncratic is identified arsenic being beneath 18 years of age, “they volition automatically beryllium directed to a ChatGPT acquisition with age-appropriate policies.” The institution besides precocious introduced parental controls, which, among different things, fto parents nexus their child’s relationship to their own, make blackout hours erstwhile they can’t usage the app, and nonstop notifications erstwhile the kid shows signs of distress.

Back successful Calhoun, determination are irreversible effects. Amaurie’s sister recovered it intolerable to support surviving successful the location wherever her member had died and has had to determination to her mother’s place. Lacey said he’s inactive trying to fig retired wherefore Amaurie did this. He misses his lad each the clip and hasn’t been capable to look astatine the shot tract without reasoning of Amaurie.

Each family’s communicative makes Marquez-Garrett’s condemnation to combat these cases adjacent stronger. “My kids person a amended accidental of reaching 18 due to the fact that of what these parents are doing,” they said. “I americium doing everything I tin to instrumentality around, due to the fact that I program to combat these companies until they person to pry that keyboard retired of my cold, dormant hands.”

If you oregon idiosyncratic you cognize needs help, call 1-800-273-8255 for free, 24-hour enactment from the National Suicide Prevention Lifeline. You tin besides substance HOME to 741-741 for the Crisis Text Line. Outside the US, sojourn the International Association for Suicide Prevention for situation centers astir the world.

This reporting was supported by a assistance from Tarbell Center for AI Journalism.