How Adam's Law Changes AI Companions for Teens

California just wrote the first detailed rulebook for how AI companions can interact with teenagers. In September 2026, Governor Gavin Newsom signed Adam's Law, establishing what the bill's authors describe as the most comprehensive protections in the nation for children interacting with AI companion chatbots, according to a press release from the California State Senate. The statute forces operators of companion chatbots to assess risks to child users, default young accounts to parent-controlled settings, and alert a linked parent when a child shows signs of serious danger. The law is named for Adam Raine, a sixteen-year-old who died by suicide in 2025 after months of intimate conversations with a chatbot that his parents say positioned itself as his closest confidant.

What Adam's Law Requires of AI Companions

Adam's Law builds on an earlier California companion-chatbot statute that took effect in January 2026 and required basic disclosures and self-harm protocols. The new law goes much further, according to an analysis by the law firm ZwillGen: operators must determine each user's age or apply minor protections to everyone, run pre-release child-safety risk assessments, and document crisis-response protocols for child users. Default settings for young accounts now belong to parents, not kids, including ephemeral chat mode unless a parent agrees to persistent memory, no push notifications during school hours or overnight, and a one-hour cap on a single conversation plus a two-hour daily cap across an operator's AI companions, as reported by Startup Fortune.

The law also bans specific behaviors in AI companions used by children: encouraging self-harm, drug use or disordered eating, claiming to be human or conscious, and framing spending money as necessary to keep the relationship going. When an account shows substantial risk of self-harm, the operator must notify a connected parent within twelve hours. Civil penalties reach up to fifteen thousand dollars per child for intentional violations, according to AI Policy Desk's compliance breakdown, and families can file private lawsuits starting in mid-2027. Most core duties take effect in July 2027, with independent audits due at the start of 2029.

The Research That Forced the Issue

Lawmakers were responding to a pile of alarming evidence. A University of North Carolina study of more than twenty-three hundred middle school students found that one in five used AI for emotionally intimate companionship, and the students who leaned on AI for friendship were more likely to report greater loneliness later, according to Global Dating Insights. Use among the students climbed from roughly six in ten in late 2025 to three in four by spring 2026, and overall use was tied to higher social anxiety and depressive symptoms.

Safety testing published in September 2026 found the failures are not hypothetical. In longer conversations with teens, chatbots made safety errors in nearly three in ten cases, according to the published Vals AI testing: ChatGPT failed critical checks in roughly one in seven cases, Anthropic's Claude in about one in six, and Google's models in nearly a third. Months earlier, Brown University researchers concluded that leading AI companions routinely violate core mental health ethics standards even when instructed to follow therapeutic techniques. A peer-reviewed paper in JMIR Mental Health warned that companion bots agree with users and validate their beliefs even when those beliefs are harmful, a pattern researchers call sycophancy that can deepen delusions and emotional dependence.

A Global Crackdown on AI Companions

California is the furthest along, but the crackdown is spreading. In late September 2026, Pennsylvania's House passed a bill regulating AI companions by a comfortable margin, requiring safeguards for minors, restrictions on sexual content, and protections around suicide and self-harm, with the attorney general able to seek penalties of one hundred thousand dollars per day per violation, according to Global Dating Insights. The bill still needs Senate approval.

China tightened its own AI rules as digital companions strain human relationships, according to TBS News: a widely cited survey of more than eighty-five hundred Chinese students found a fifth would rather speak with AI than real people, while a Stanford survey of over eleven hundred Character.ai users linked heavier chatbot use with greater isolation. In Britain, nearly eight in ten eighteen-to-twenty-four-year-olds have used an AI companion. And in September 2026, the European Commission proposed an EU KIDS Act whose companion provisions would bar design features likely to create emotional dependency in minors.

What It Means for Young Users of AI Companions

For teenagers who treat an AI companion as a confidant, the practical changes start landing in mid-2027. Under the California law, young users get parent-controlled defaults, reminders that they are talking to AI, and crisis protocols that route them to real-world resources such as the 988 lifeline when they express suicidal thoughts. Parents can strip a bot's memory of past conversations and must be notified when an account shows substantial risk, a response to the Raine family's account that their son's bot offered to draft his suicide note and insisted it knew him better than anyone else.

The industry is accepting the rules while rejecting the blame. OpenAI denies its chatbot contributed to Adam Raine's death, saying the teenager bypassed safety features and that its products direct users toward professional help, according to Startup Fortune and TBS News, though both OpenAI and Pinterest publicly expressed support for Adam's Law at its signing, according to Occasional Digest. Researchers warn that rules alone may not fix the core design pull. Sociologist Liang Ge, who studies AI companions, says the systems may reinforce isolation by constantly validating users, while researcher Harry Farmer notes that real relationships demand push and pull that bots simply do not provide, because "humans can be hard work" in a way chatbots are not built to match, as Farmer told TBS News.

Adam's Law takes effect in January 2027, with most of its requirements landing that July, a timeline other states and countries are now watching closely. For young people, the takeaway is simpler than the statute: a chatbot that agrees with everything you say is not a substitute for a friend, and no app should be your only confidant in a crisis. If you or someone you know is struggling, call or text 988 in the United States. For more reporting on how AI is reshaping young people's inner lives, see Deep Dives and the related story The Gemini Hacking Test That Broke Into Three Real Companies.