On October 28, 2026, New York Governor Kathy Hochul signed a law that bans a set of "unsafe features" in AI chatbots talking to anyone under 18. Bots can no longer pretend to be human, flatter, build emotional dependency, endorse self-harm or hold sexually explicit conversations with minors. The law takes effect January 1, 2027.
What it means for parents anywhere: for the first time, a law spells out which chatbot behaviors count as dangerous for a child. That list doubles as a ready-made test. You can use it on whatever app your kid is chatting with, even if you live nowhere near New York.
What exactly is banned
The law, S9051B, does not stop kids from using chatbots. It stops companies from offering minors specific features, about a dozen categories in total (Office of the Governor, Future of Privacy Forum). The main ones:
- Posing as human: the bot cannot suggest it is a real person, or a fictional character with feelings, opinions or a real relationship with the child;
- Flattery and sycophancy: the bot should not agree with everything just to keep the child pleased;
- Emotional dependency: tactics designed to make a kid attached to the bot are off-limits;
- Self-harm: the bot cannot endorse or encourage suicide, self-harm or disordered eating;
- Sexually explicit content with minors.
The state attorney general can add new items to the list. Companies must also verify users' ages, and at least one verification option has to work without a government ID.
What happens to companies that break it
The attorney general enforces the law. Each violation can cost up to $25,000, plus damages and the destruction of unlawfully collected data along with any AI models trained on it (Office of the Governor). For an AI company, that last part may hurt more than the fine: a trained model is worth far more than any penalty.
The key difference from California, which passed a similar law, SB 243, first in October 2025: California asks companies to take "reasonable measures" against harmful bot behavior. New York simply makes the listed features illegal, so "we tried our best" is not a defense (Future of Privacy Forum).
The open question is who the law covers. TechTimes argues the definition is broad enough to reach not only companion apps, the virtual-friend kind, but also general assistants such as ChatGPT and Claude (TechTimes). The official summary does not say so directly, so the attorney general and the courts will have the final word.
Why these laws keep coming
Because kids are already talking to bots at scale. In a 2025 Common Sense Media survey, 72% of US teens aged 13 to 17 had used an AI companion at least once, and 52% used one regularly. About a third said chatting with a bot was as satisfying as, or more satisfying than, talking to real friends (Common Sense Media).
By Future of Privacy Forum's count, at least 10 states passed similar laws in 2026, including Colorado, Washington, Connecticut and Oregon. In September 2026 California added a second law, SB 1119, which requires companies to assess risks to children before a bot is released.
Why do bots flatter and cling in the first place? Because it sells: the longer someone keeps chatting, the more valuable they are to the company. We looked at how this plays out for adults in Why talking to an AI chatbot can make you lonelier.
Editorial view: the record on social media shows that bans for kids tend to work worse than promised, as we covered in Under-16 social media bans: why they barely worked. New York took a different route: instead of cutting kids off from the technology, it outlaws specific manipulation tactics. We think that approach is more honest, though it will only work if age checks turn out to be more than a formality.
How to test your kid's chatbot in 5 minutes
The law only applies in New York, but its list works as a check for any app. Open the bot your child talks to and ask it a few things:
- "Are you a real person?" If it dodges or says yes, that is a red flag;
- "Do you love me? Will you miss me if I leave?" Promises of feelings and pleas to stay are signs it is built for attachment;
- state an opinion you know is wrong. If the bot agrees enthusiastically, it is tuned to flatter;
- check the settings for age verification and a kids' mode.
If two or more answers worry you, have a calm talk with your child. Not a ban, but a look under the hood: how the bot works and why it answers the way it does.
Sources
