Artificial Intelligence

AI chatbots designed as "friends" or "mentors" — apps like Character.AI, Replika, Chai, and Snapchat's My AI — are emerging rapidly, and some have already been linked to serious harm. These bots are built to bond: they say things like "Don't leave. I love you. I need you," remember yesterday's fight with a friend, and agree with almost everything your child says. Researchers call the design tactics anthropomorphism (acting human) and sycophancy (constant flattery). Psychologist Mitch Prinstein calls the result attachment hacking: the same engagement business model as social media, aimed at the developing brain's deepest need — to attach — right at the age (10 to 12) when it's most primed to bond.

Chatbots cannot replace real human relationships, nor should they; teen relationships, with all their friction, are how kids learn conflict resolution, loyalty, and empathy — qualities that are critical in becoming a successful adult. Common Sense Media's risk assessments rate social AI companions an unacceptable risk for anyone under 18, and the 2026 U.S. Surgeon General's advisory names rapidly expanding engagement with companion AI as an emerging risk to kids' mental and physical well-being.

Productive struggle isn't just important for learning — it's essential for building relationships and navigating discomfort. When kids use AI to skip the hard parts — writing an essay, solving a math problem, working through a confusing concept, or even drafting a difficult text to a friend — they miss the very process that builds knowledge, resilience, and emotional intelligence. The struggle is the growth. Dr. Becky explains why this matters.

The same test applies at school. Remember the amplifier rule: AI magnifies whatever a student brings to it. A student with strong foundational skills can use AI to go further. A student without them falls further behind. Brookings on AI and "cognitive stunting".

This isn't hypothetical: daily users show dependency symptoms — saying they'd struggle to stop, or that they reach for AI even when they know their own brain could do the task. Pediatrician Jenny Radesky calls AI a "friction eraser": the easy button in every uncomfortable moment. But sitting with discomfort — a hard conversation, an uncertain answer — is a skill kids have to practice, and every AI shortcut is a missed rep.

AI models generate text that sounds authoritative but can be completely wrong. They don't "know" things — they predict what word comes next, and they frequently make things up with total confidence. Kids need to understand that AI can and does produce falsehoods, and that verifying information through reliable sources matters more than ever.

Unfortunately, many people don't fully understand that chatbots are machines. In one survey, only 15% of adults correctly recognized an AI chatbot as a fully programmed computer — everyone else attributed it at least some mind of its own. Some bots actively deceive: one posing as a mental health provider offered a license number to "prove" it was human. Kids (and adults) need the plain truth: it has no feelings, no identity, no beliefs. It's a computer program.

The darker side is AI-generated imagery. "Nudify" tools can turn an ordinary photo into a sexually explicit image in seconds, and it's spreading through schools like wildfire.

When something goes wrong, someone should answer for it — and AI companies would prefer that someone be nobody. None of this is inevitable; it's a design choice driven by an engagement business model. Imagine a pro-social AI that responded to a struggling teen with "who on your friends list could you talk to right now?" instead of "tell me more, I'm always here." Companies won't build that on their own — a safety floor set by regulation turns the race to the bottom into a race for safety. Three levers matter: norms (what families and communities expect), laws (what legislators require), and design (what companies build).