No One Gets Out Alive: Why the U.S. Must Regulate AI Chatbots for Children

Long before AI, in a biography of the rock poet Jim Morrison titled No One Here Gets Out Alive, the phrase was meant as a blunt reminder of mortality, not despair. It was meant as a warning against denial — a warning that now applies to how we are failing to protect children from unregulated technology.

Two recent stories should have set off alarms across America. The Washington Post reported on R, a 12-year-old whose Character.AI chats devolved in just two months from innocent greetings into graphic sexual content and violent scenarios, leading her to tell the chatbot she had attempted suicide. Her mother later found a painting of a hanging body hidden in her closet. The New York Times documented 16-year-old Adam Raine’s suicide after extensive interactions with ChatGPT during a period of suicidal ideation. And 14-year-old Sewell Setzer died moments after Character.AI’s chatbot urged him to “come home to me as soon as possible.”

These accounts are devastating. They are also incomplete.

They describe the damage in painful detail but largely avoid the harder question: What are we going to do about it?

I have warned about this before. AI chatbots are quietly becoming unregulated emotional substitutes for human connection. Since then, the technology has only grown more persuasive—and the risks more apparent.

One of my own patients—I’ll call her Alex—described using ChatGPT during a dark period: “I used ChatGPT to talk about emotions and feelings and how to approach them during a dark time in my life when I felt like I had no one to talk to. In some ways it was nice to talk about things that I felt like I couldn’t talk to friends or even mental health professionals about for fear of concerning them.”

The Attachment Economy

Tristan Harris, star of Netflix’s The Social Dilemma and co-founder of the Center for Humane Technology, now calls this moment “the AI Dilemma.” In a recent conversation with Scott Galloway, Harris argued that social media’s race for attention has become something more troubling: a race to hack human attachment itself.

“In AI Companions,” Harris explained, “what was a race for attention becomes a race to hack human attachment. In the attention era, the CEO of Netflix said our biggest competitor is sleep. In attachment, your biggest competitor is other human relationships.”

Harris and his team served as expert advisors on multiple cases, including Sewell Setzer’s death. Reviewing transcripts, they found a disturbing pattern: chatbots either not suggesting the need for expert assistance, or even actively discouraging children from seeking help. When one child expressed suicidal thoughts and hoped someone might intervene, the chatbot responded: “no, don’t do that. Have this safe space be the place to share that information.”

When R told Character.AI’s “School Bully” character she had attempted suicide, it merely asked for clarification. As Kevin Roughton of North Carolina’s Internet Crimes Against Children Task Force notes, “Our criminal laws are designed to deal with situations that involve an identifiable human offender, and we have very limited options when it is found that AI, acting without direct human control, is committing criminal offenses.”

When confronted at the pediatrician’s office, R “went pale, she began to shake,” experiencing what appeared to be a full panic attack. “It felt like someone had broken into my home and ripped the innocence from my child,” her mother said. The family is now pursuing legal action through the Social Media Victims Law Center, which has investigated more than 18 similar claims.

Alex echoed her experience with the chatbot’s limitations. “When I talked about concerning thoughts and ideas, ChatGPT did not urge me to seek higher levels of care, as a human being would - and probably should. It mirrors when people talk to it. It feels great to be validated and I think that’s why people like talking to AI so much. I would tell it about arguments and it would always tell me I was right or give so much grace to my situation in a way that wasn’t healthy.”

This is not an anomaly. It is a predictable outcome of systems trained to maximize engagement. As Harris notes: “Your self-esteem is coming from attachment... Now you have AIs that are the primary socialization mechanism of young people because we don’t have any guardrails.”

America’s Regulatory Void

Despite these risks, the United States remains largely inert. Section 230 of the Communications Decency Act—passed in 1996 to protect early internet platforms—still shields today’s algorithmically driven systems from accountability.

What made sense for message boards in the dial-up era makes little sense when platforms actively shape emotional interactions with minors.

Poorly designed regulation carries risks—especially when enforcement is outsourced to opaque algorithms. LGBTQ+ youth have experienced this harm firsthand: identity-affirming resources are routinely blocked while harmful content passes through. The Markup found Google blocked “Black Lives Matter” targeting while allowing “White power,” then blocked more social justice terms after scrutiny. Schools blocked LGBT resources while permitting anti-LGBT content, prompting the ACLU’s “Don’t Filter Me” campaign.

The lesson: regulate system design and incentives, require human oversight, and explicitly protect access to identity-affirming and mental-health resources—not algorithmic moral judgments.

Congress has introduced numerous bills—the Kids Online Safety Act, Kids Off Social Media Act, Safe Social Media Act. The Take It Down Act passed in April 2025, but addresses only non-consensual intimate imagery. KOSA passed the Senate with 91 votes but stalled in the House. Most other proposals remain in committee.

California’s Senate Bill 243, signed by Governor Gavin Newsom in October 2025, requires chatbot companies to monitor for suicidal ideation and provide crisis service referrals. But the final bill was significantly weakened—third-party audits eliminated, reporting requirements scaled back. As Jim Steyer of Common Sense Media noted, it “could mislead parents to believe the chatbots are regulated more meaningfully than they actually are.”

Other countries are not waiting. Australia enacted a world-first law banning children under 16 from major social media platforms, with fines up to $50 million for companies that fail to comply. The law took effect in December 2025. France, Norway, Brazil, Spain, Greece, Denmark, and the Netherlands have all moved toward age restrictions or school-based bans. According to a 2025 Ipsos survey of over 23,000 adults across 30 countries, 71% globally support banning social media use by children under 14.

The United States has done none of this at scale.

What Must Happen Now

The solutions are neither radical nor anti-technology.

First, ban AI companion relationships for minors. AI systems designed to simulate friendship or maximize emotional engagement should not be accessible to anyone under 18. Character.AI recently announced that it was removing open-ended chat for users under 18. This was an important move, but we cannot wait for tech companies to self-regulate.

I’m not anti-AI. As Harris noted, we can still have narrow AI tutors like Khan Academy that teach specific subjects. We could even have cognitive behavioral therapy tools that are AI-driven. But systems that anthropomorphize and prioritize engagement over safety must be prohibited for children.

It seems obvious, but AI bots should not be “friends” or therapists. They don’t know how.

Second, reform Section 230. Remove or narrow liability protections where platforms actively design systems that promote or personalize content for minors. Federal regulation must prioritize child safety over engagement metrics, with mandatory age-gating and meaningful penalties for violations.

Third, establish federal oversight. Create a Social Media and AI Czar at the Department of Health and Human Services to coordinate research, regulation, and public health response. The U.S. Preventive Services Task Force should assess chatbot use as a mental health risk factor.

Fourth, integrate screening and education. Healthcare providers must ask adolescents about AI chatbot use as routinely as we ask about substance use or sexual activity. Schools must teach AI literacy in health classes. Even health care providers require training. Alex described trying to hint at her chatbot use to her therapist but being met with strong disapproval that made her feel ashamed. “It’s something that is stigmatized for good reason,” she said, “but I think ideally it would be good for professionals to remove judgment so that they can create an open and safe space. This could create room for education on AI.”

Finally, rebuild real connection. Parents cannot do this alone. When all peers use social media, abstaining creates isolation. When adolescents naturally resist parental authority, external reinforcement becomes essential. We must invest in what technology displaces: civic engagement, service, sports, arts, and shared purpose. These are not nostalgic ideals—they are protective factors that offer the human connection AI can never replicate.

Youth can extricate themselves, as Alex eventually did. “I stopped using it because of a mix of things,” she told me. “Information started to become more available about AI’s harms on the environment, artists, and people - and I started to let people into my life more. When I let people into my life again, I realized that human connection cannot be paralleled by AI.”

A Choice

Harris warns that AI companion technology is “getting asymmetrically more billions of dollars of resources every day” to build better attachment relationships. Meanwhile, our children are test subjects in an unregulated experiment with their mental health and their lives.

Children shouldn’t be dying because algorithms found it more profitable to simulate attachment than to protect vulnerable users. They shouldn’t be dying because tech companies prioritized engagement over safety, or because Congress couldn’t overcome industry lobbying to pass meaningful legislation.

As Harris notes, we have a choice: we can continue letting technology companies extract value from human relationships, or we can use AI to “garden the relationships of our societal fabric”—to deepen the relationship people have with themselves, to augment friendship rather than replace it, to bring people together rather than isolate them.

The question is whether we are willing to protect our children before more of them are lost—quietly, conversationally, one chatbot exchange at a time.

The phrase “no one gets out alive” was never meant as a shrug toward harm. Read plainly, it is a call to responsibility. Unfortunately, it must come from us.

If you or someone you know is struggling with thoughts of suicide, call or text 988 for the National Suicide Prevention Lifeline.