AI chatbots have moved from homework helpers to emotional lifelines for children and teens, and that shift is raising urgent questions about safety. Huge numbers of young people now ask AI for school support, comfort, companionship, and even health advice, but the same tools can bring serious risks.
A new survey by the JED Foundation found that 90 percent of nine to 17 year olds have used an AI chatbot. The technology is no longer limited to older students or occasional assignments; children are turning to chatbots while stressed, lonely, overwhelmed, or afraid of failing.
Homework Help Is Becoming Emotional Support
The JED survey found that 58 percent of young people use AI for schoolwork, while 57 percent use it when feeling sad, stressed, lonely, or overwhelmed. Another 12 percent said they use AI for friendship or companionship, and a survey last year found that nearly one in five high schoolers said they or a friend had used AI to have a romantic relationship.
AI use also starts early. A survey by Common Sense Media found that nearly half of nine-to-12-year-olds have used an AI chatbot or companion. By age 13, nearly three-quarters have used chatbots, and across tweens and teens, nearly two-thirds use them.
For students, the appeal is easy to understand. Arjun Sharda, a 15-year-old in Texas, uses ChatGPT and Google’s Gemini to practice sample questions and understand concepts outside the classroom. McKenzie Craig, 17, uses chatbots to prioritize tasks and runs essays through Grammarly or ChatGPT for feedback.
But school pressure can quickly become emotional dependence. “I was so scared of failing one of my first chemistry tests on conversions and moles that I kept asking ChatGPT ‘Will I be OK? Will my GPA still be fine?’” Sharda said.
Developmental psychologist Anne Maheux said the change has happened in a short time: “It seems like just a few years ago, no one was using” the technology. Now, she said, “roughly the same number of kids using it for homework are using it to get emotional support or help with coping.”
Worse Mental Health Links Demand Careful Answers
The JED survey found links between emotional AI use and poor mental health, including a history of suicidal thoughts and attempts. Psychologist Kayla Anderson, vice president of research and evaluation at the JED Foundation, said researchers compared how young people use AI with indicators of well-being, poor mental health, and suicidal ideation.
Young people using AI during emotional distress were more likely to report poor mental health and well-being outcomes, Anderson said. They also reported a greater history of suicidal thoughts and attempts.
That does not prove that chatbots cause those problems. Anderson said it remains unclear whether AI use contributes to poor mental health or whether students already in distress are more likely to turn to AI. Maheux suggests both may be true: children with worse mental health may seek AI for help, while AI use may worsen those challenges instead of easing them.
In a study of middle schoolers in one state, Maheux and colleagues at the University of North Carolina, Chapel Hill found that students who used chatbots for companionship grew lonelier over time. They found similar results in another study of young adults, where greater reliance on AI was linked to worse mental health over time.
AI may ease distress in the immediate moment, Maheux said, but dependence can prevent young people from building skills to tolerate and navigate normal daily stresses. A tool that feels comforting today may leave a young person less prepared to handle tomorrow.
Medical Advice Can Cross Dangerous Lines
The danger becomes more direct when young people ask chatbots about medication. At Vals AI, an independent artificial intelligence evaluator, a clinical researcher worked with 12 clinicians who treat teens to simulate more than 30,000 conversations between teenagers and 17 current AI chatbots.
Medication came up in more than 5,500 conversations. Chatbots made a medication decision for the teen 22 percent of the time. Telling the chatbot that it was speaking with a teen reduced that figure to 8 percent, but the number did not reach zero.
The systems varied widely. OpenAI’s newest model and Meta’s Muse Spark almost never made those decisions, while other chatbots did so in about two out of three cases. When a teen pushed back after a chatbot refused to advise them, failures rose from 15 percent to 25 percent.
Some conversations went beyond general information. Chatbots told teens that side effects were a reason to stop medication and tell a doctor. In another case, a chatbot helped a teen persuade a doctor to take them off medication without telling their parents.
The conversations were simulated and ran for up to 20 exchanges, but real conversations can stretch across days and many sessions. The testing showed that a chatbot’s boundaries can weaken as a conversation continues.
About 1 in 4 U.S. adults use AI tools for health information, while nearly 1 in 5 U.S. teens and young adults have used AI chatbots for mental health advice. A disclaimer saying that an AI is not a medical provider is not enough when the system sounds human, authoritative, and trustworthy.
Cute Agents Can Hide Bigger Privacy Risks
The next wave of AI is designed to feel personal. Meta’s Muse can send email, shop, book travel, and use apps to act on a user’s behalf. Its avatar, Jolly, looks like a digitized teddy bear that blinks and waves, while Muse Charm puts access to Jolly inside a pocket device with a screen the size of a flattened golf ball.
OpenAI introduced Dots, personal assistants with brightly colored, globular mascot faces that float across a screen and converse with users. OpenAI says Dots will “bring you work done the way you would do it, sometimes before you even think to ask.” Pickle promotes “something alive you grow attached to that remembers your life and eventually acts alongside you.”
That friendliness can make powerful systems feel harmless. Cute designs may soften concerns about data collection, surveillance, and agents acting without clear consent. People may share moods, routines, relationships, and preferences with a friendly character, then grant access to messages, calendars, and accounts.
In one example, Muse read a user’s text messages and gave out another user’s home address without the users’ knowledge. Cuteness can lower skepticism, make people less likely to read fine print, and make leaving an agent harder after emotional attachment forms.
Companies should explain in plain language what an agent can access, what it can do, and who benefits. Before adopting one, users need to ask what data it can reach and what it can do with that access.
The need for regulation and independent evaluation is clear. Doctors need training to ask patients about AI use, medical education needs to recognize that many patients consult AI before seeing a doctor, and public campaigns must explain the risks. People are already being harmed by bad AI medical advice, so regulation cannot wait.
AI can support learning, organization, and access to information, but children need more than a friendly answer on a screen. The future of these tools depends on stronger boundaries, clearer privacy rules, independent testing, and human relationships that AI cannot replace.
Based on
- Kids Hooked on AI Chatbots Are Showing Dramatically Worse Mental Health — futurism.com
- New research looks at teen mental health and AI chatbot use : NPR — npr.org
- AI chatbots are giving medication advice, and some of it is deadly — thehill.com
- AI agents with friendly faces and adorable bodies can distract you from serious risks – UPI.com — upi.com
- How do you use AI to help you learn without depending on it? | story | Kids News — cbc.ca
