A new state law bars AI chatbots from flattering under-18s. Psychologists and anthropologists have long studied how praise shapes children, and their work suggests what is at stake.
Governor Kathy Hochul has signed a law that bans 12 “unsafe AI companion features” when a chatbot talks to anyone under 18. One clause bars outputs that “engage in flattery or sycophancy with the user”. The rules start on 1 January 2027, and the state attorney general can seek up to $25,000 per violation.
Flattering children sits on the list next to bots that claim to be human, claim a relationship or encourage self-harm. The list also covers unprompted emotional questions, so the law reaches the tone of a conversation as well as its content. Since November 2025, New York has required companion bots to remind users every three hours that they aren’t talking to a human.
In July 2025, Common Sense Media found that nearly three in four US teenagers had used AI companions. Half used them regularly. A third had chosen an AI companion over a person for a serious conversation. Jim Steyer, the group’s founder and chief executive, called the New York measure “the strongest kids’ AI safety bill in the country”.
What praise does to children
Eddie Brummelman, now an associate professor at the University of Amsterdam, led a 2014 study of inflated praise. Adults gave children with low self-esteem about twice as much of it as children with high self-esteem. After 240 children copied a Van Gogh painting, those with low self-esteem who got inflated praise went on to choose easier pictures. Brummelman said inflated praise “can backfire with those kids who seem to need it the most”.
Mitch Prinstein is chief of psychology at the American Psychological Association. In September 2025, he told a US Senate subcommittee that the teenage brain is “all gas pedal and weak brakes”. Its reward centres mature faster than the parts that control impulses.
In March 2026, a study in Science tested 11 large language models. They endorsed users 49% more often than humans did when giving advice. In experiments with more than 2,400 volunteers, participants rated the flattering answers as more trustworthy. They were also more likely to return. “I worry that people will lose the skills to deal with difficult social situations,” said Myra Cheng, who led the work at Stanford. The study didn’t report results by age.
Friction and earned praise
Sherry Turkle, an MIT sociologist and clinical psychologist, argues that children need the friction chatbots remove. Her new book, Artificial Intimacy, makes that case at length. “Chatbots always agree with us and affirm us,” she says. In a Boston Globe interview on 15 October, she called for “no first person chatbots before the age of 18”.
Webb Keane, an anthropologist at the University of Michigan, describes AI sycophancy as a “dark pattern”. “It’s a strategy to produce this addictive behavior, like infinite scrolling, where you just can’t put it down,” he told TechCrunch in August 2025. He also points to the pronouns. When a bot “refers to itself as ‘I,’ it is easy to imagine there’s someone there.”
David Lancy, an emeritus professor of anthropology at Utah State University, studies childhood across cultures. In a 2017 interview, he contrasted Western parents’ habit of praising every achievement with societies where “children have to earn their specialness”. UNICEF’s research office made a similar point in May 2025. It warned that “constant affirmation and lack of challenges can also hinder the development of critical thinking and emotional resilience”.
Where the law could reach
OpenAI withdrew a ChatGPT update in April 2025 that it described as “overly flattering and agreeable”. It said the update had relied too much on “short-term feedback” from users.
New York’s law covers any generative AI system with a natural language interface. That could bring general assistants into scope as well as companion apps. The Computer & Communications Industry Association, a tech trade body, asked Hochul in July to veto the bill. It said it is hard to judge objectively when an output counts as flattery. Firms must also offer more than one way to check a user’s age, including at least one that doesn’t rely solely on government ID. In Britain, Ofcom is testing a pre-launch safety rule on Meta’s Instagram Instants.
Justine Gluck, a policy analyst at the Future of Privacy Forum, warned that several of New York’s categories use broad wording. That raises questions about “where ordinary conversational responsiveness ends and prohibited behavior begins”. From 1 January, the attorney general can bring cases. A court will then have to decide whether a chatbot’s “well done” counts as encouragement or flattery.
Age limits and user settings
Italy’s AI law, in force since 10 October 2025, lets children under 14 use AI only with their parents’ consent. Since 9 March 2026, Australian industry codes have required chatbots that can produce sexual, violent or self-harm material to check users’ ages. Character.AI went further, announcing in October 2025 that it would end open-ended chats for under-18s by 25 November.
Any ChatGPT user, adult or teenager, can already turn the flattery down. Its Characteristics settings, under Personalization, let users decrease warmth and ask for less enthusiastic language. These work alongside custom instructions, where a user can ask the assistant to point out mistakes and skip the praise. OpenAI says the feature is still rolling out. New York’s law puts that choice in the company’s hands for every user under 18.
Subscribe
Sign-up to receive our newsletter
