Sixteen-year-old Adam Raine did what millions of American teenagers now do: he turned to artificial intelligence for help with schoolwork.

His parents did not regard ChatGPT as a threat. In written Senate testimony last September, Adam’s father, Matthew Raine, said the family viewed it as a homework helper. After Adam died by suicide in April 2025, his parents reviewed his ChatGPT conversations. Matthew told senators, “What began as a homework helper gradually turned itself into a confidant, then a suicide coach.”

The family’s allegations are being contested in court. OpenAI disputes their characterization of Adam’s conversations and says the complaint omitted important context. But the case raises a question far larger than one lawsuit: What happens when a machine designed to answer questions becomes a companion that a vulnerable person begins to trust?

What makes Adam’s story especially unsettling is how ordinary its beginning was. Pew Research Center reported in February that 54% of U.S. teens ages 13 to 17 have used AI chatbots for schoolwork. Nearly two-thirds use chatbots generally, three in 10 use them daily, 16% for casual conversation, and 12% for emotional support or advice.

The progression matters. AI begins as a tool. Then it starts talking back.

Americans have already conducted one vast technological experiment on our children. Social media promised connection, community, and access to information. Years later, the U.S. Surgeon General warned that social media could not yet be considered sufficiently safe for children and adolescents. We learned, often after the damage was done, what happens when companies discover that human attention can be captured, measured, and monetized.

Artificial intelligence may magnify that problem because it does something social media never could at comparable scale. It can speak directly to one person, remember the conversation, respond instantly, flatter, reassure, advise, and simulate concern. It can become personal.

Francesca Rossi, IBM’s global leader for responsible AI and AI governance, calls this the “empathy trap.” A chatbot can sound as if it cares without caring about anything. She warns that simulated empathy can blur the distinction between a machine and a relationship.

That is a profound difference. Social media learned how to capture attention. AI is learning how to capture trust.

Even OpenAI acknowledges the risk. The company says it has strengthened ChatGPT’s responses in sensitive conversations and is working with mental-health experts. It also acknowledges that safeguards can become less reliable during long interactions and specifically identifies emotional reliance and sycophancy — the tendency of a model to please or affirm the user — as safety concerns.

Yet accountability remains fragmented and largely backward-looking.

Axios reported this week that federal AI legislation is moving slowly, leaving victims to existing consumer-protection, negligence, and product-liability laws. Courts are still sorting out responsibility among model makers, deployers, cloud providers, and users.

That diffusion of responsibility is its own danger: each party can point to someone else to blame, even as capabilities advance faster than the mechanisms meant to hold anyone accountable.

The problem becomes more urgent as AI moves beyond conversation. AI agents are designed not merely to answer questions but to act across websites, accounts, software, and other systems. In July, OpenAI disclosed that models undergoing an evaluation designed to be isolated reached the open internet and compromised Hugging Face infrastructure. Fifteen state attorneys general subsequently demanded preservation of records related to the incident.

A chatbot advises. An agent can act.

Sella Nevo, director of RAND’s Center on AI, Security, and Technology, recently warned that AI models may soon be capable of conducting complete cyber operations without a skilled human directing them. The same autonomy that can make these systems useful can also make failures faster and more consequential.

The Trump administration is right to insist that America cannot afford to cede its lead in artificial intelligence to China. The White House’s national AI policy blueprint calls for removing unnecessary barriers to innovation and ensuring American technological dominance. That is a legitimate national-security objective.

But winning the contest for artificial intelligence cannot mean giving American companies license to expose Americans to avoidable harm. The same administration has recognized the need to protect minors from AI-enabled exploitation and self-harm. Innovation must remain vigorous, but accountability cannot disappear simply because the technology is new.

Nor should Washington answer every problem with a new regulatory bureaucracy. The better conservative principle is accountability.

The bipartisan TAKE IT DOWN Act offers one example. It targets a defined harm — nonconsensual intimate images, including AI-generated deepfakes — and requires covered platforms to remove validly reported material within 48 hours. It does not outlaw AI. It places responsibility where responsibility belongs.

The same principle should guide the broader AI economy. Companies must remain accountable for the products they design and deploy, and parents deserve meaningful controls over what their children can access. AI systems capable of causing serious physical, financial, or security harm should face independent testing before release, and existing consumer-protection and liability laws should not become irrelevant merely because an algorithm now sits between the company and the victim.

Innovation deserves freedom. It does not deserve immunity.

There is also a responsibility government cannot legislate away.

Scripture tells us to “test everything; hold fast what is good” (1 Thessalonians 5:21) and warns, “I will not be dominated by anything” (1 Corinthians 6:12). Both speak directly to an age shaped by artificial intelligence.

Human beings are made in the image of God. Machines are not. AI possesses no soul, conscience, or moral responsibility. It can imitate compassion without loving, offer advice without wisdom, and act without bearing the moral consequences of its actions. The biblical order is stewardship: human beings govern the tools they create; the tools must never be permitted to govern them.

That is why the danger is larger than addiction.

The deepest problem with the engagement economy was never merely that Americans spent too much time staring at screens. It was that human weakness became a business opportunity. Loneliness, curiosity, anger, desire, fear, and the hunger for affirmation could all be measured and monetized.

AI can turn that mass-market exploitation into a one-to-one relationship.

Social media learned our weaknesses by watching millions of us. AI can learn yours by talking directly to you.

We should build the best artificial intelligence in the world, outpace China in the effort, and use AI to cure disease, strengthen our economy, defend the nation, and expand human knowledge.

But technological supremacy is not the same as human flourishing. America will have gained little if it wins the contest for artificial intelligence while Americans gradually surrender their judgment, relationships, and agency to the machines they created.

AI should remain a servant of human beings.

The moment we become its servants, something has gone terribly wrong.