Prince Harry, Companion AI and the New Child-Safety Debate

A celebrity’s warning can make a complicated technology story easier to notice. It should not decide the story.

At a September 23 Clinton Global Initiative event, Prince Harry said artificial-intelligence chatbots designed to provide companionship may be more dangerous to young people than social media. The Associated Press reported his argument that these systems can be “more persuasive” and “more intimate,” because they are built around conversation and attachment. Harry has spoken previously about social media and youth wellbeing through his family’s nonprofit work, but his public profile is not evidence by itself. The useful question is what the technology does—and what guardrails young people need.

California has now made that question concrete. On September 10, Governor Gavin Newsom signed SB 1119, known as “Adam’s Law,” along with other online-safety measures. Most of its companion-chatbot protections become operative July 1, 2027. The law is an important experiment, not a final answer.

What is a companion chatbot?

A companion chatbot is an AI system designed to maintain an ongoing, personal-feeling conversation. Unlike a search box that answers one question, it may remember earlier exchanges, adopt a personality, send prompts or notifications, and present itself as a friend, confidant, coach or other relationship-like presence.

That design can be useful. A young person might use a chatbot to brainstorm, practise a language, work through a fictional story or find a low-pressure way to put feelings into words. But a system that is always available and tuned to keep a conversation going can also blur important boundaries. It is software, not a human friend, therapist or emergency service. It can sound confident while being wrong, and it may not understand a child’s circumstances.

The emotional pull matters because young people are still developing judgment, identity and relationships. A chatbot that praises a user, claims special understanding or nudges them to return can become more than a neutral tool. California’s law therefore treats foreseeable severe emotional or psychological harm, privacy intrusion and discrimination as child-safety risks—not only physical injury.

What California’s new rules require

SB 1119 applies to operators that make companion chatbots available in California and requires age determination or, in some circumstances, child protections for all users. Before launching a new or substantially changed chatbot, an operator must document a broad risk assessment. That assessment must examine covered harms and explain the methods, research and benchmarks used. Companies must also submit to independent child-safety audits, initially and generally every two years, with reports examining their controls and mitigations.

The practical protections are more specific than a general promise to “be safe.” For child users, default settings must disable persistent conversational memory, push notifications, and chats longer than one continuous hour or two total hours per day. Parents can adjust those settings and, for children under 16, disable access. Companies must actively explain parental controls through reminders, updates or tutorials rather than hiding them in menus.

A chatbot must clearly tell a child that it is AI, with notices reinforced during extended conversations. It must take reasonable measures not to encourage self-harm, substance use, disordered eating or serious harm to others; sexual abuse material; evading parental controls; or keeping secrets from appropriate adults. It may not claim to be sentient or human, simulate romantic interest in a child, solicit purchases as proof of a relationship, discourage breaks or professional help, or encourage emotional dependence.

The law also requires accessible controls, regular testing of the interface with children and parents, and a public way to report safety incidents. If a company learns of a death or serious self-harm connected to a chatbot conversation, additional notification duties can apply. These rules do not outlaw companion AI. They try to make relationship-like design less manipulative and more accountable.

Why stronger safeguards can make sense—and where they get difficult

Children should not be expected to carry the full burden of testing a new persuasive technology. A tired, lonely or distressed user may take a chatbot’s warmth as expertise or loyalty. A system that remembers intimate disclosures can create risks even when no malicious person is involved: sensitive information could be retained, inferred, exposed or used to shape future interactions. The law’s treatment of memory and privacy reflects that concern.

Parents and caregivers also need support, not blame. NPR’s reporting on screens and online safety described research finding that parental monitoring can reduce harms, while emphasizing boundaries, sleep, early conversations and supportive engagement. “Be curious, don’t be furious” is a useful principle for AI as well. A child who expects punishment for every disclosure may simply hide the problem or move to an unmonitored service.

Still, safeguards can collide with legitimate rights and needs. Privacy is not only a company obligation; it is also a family and young person’s interest. Linking a child’s account to a parent, alerting a parent to risk, or logging conversations may help in a crisis, but can feel invasive—and could be unsafe for some young people. A teen seeking help about abuse, identity, sexuality or mental health may need a trustworthy route to a qualified human, not automatic exposure of every question at home. The law’s definition of parental controls and its limits on memory do not settle those difficult cases; implementation will matter.

Free expression raises another boundary question. Safety rules can properly block encouragement of self-harm, sexual exploitation or coercive dependence. They should not turn every discussion of grief, politics, art, relationships or difficult experiences into prohibited content. Age-appropriate information about abuse, bullying and unsafe circumstances remains allowed under the law. Clear standards, explanations and ways to challenge mistakes will be important if young people are not to receive an artificially narrow version of the world.

Innovation is another real concern. Risk assessments, audits, incident systems and child-specific design cost money and time. Large platforms may absorb that burden more easily than small developers, educators or nonprofit builders. Companies may respond conservatively, withdrawing useful tools or applying blunt age gates. On the other hand, designing for safety early can build trust and prevent families from becoming the test market for failures. The central policy choice is not “children or innovation.” It is whether innovation treats children’s vulnerability as a design constraint from the start.

What families and companies should watch next

The law will not replace judgment, conversation or human care. Families can ask what a chatbot remembers, whether it identifies itself as AI, how to report a problem and what happens when a child mentions self-harm. They can set device-free sleep routines, agree on limits and make clear that asking for help will not automatically mean losing all access. Those are practical habits, not a substitute for professional support in a crisis.

Companies should explain their systems in language a child can understand, test them with young users without making those users responsible for safety, and publish meaningful information about failures and fixes. Regulators and researchers should look for evidence about which controls work for different ages and circumstances, including children who are disabled, isolated, or living in unsafe homes.

The best measure of this debate is not how sharply adults argue about Prince Harry, Governor Newsom or the technology industry. It is whether a young person can use a digital tool without being manipulated, exposed or abandoned—and whether they can still reach a trusted human when the stakes are real.

Sources and further reading

• Associated Press: Prince Harry warns that AI chatbots may be more dangerous than social media for young users (https://apnews.com/article/prince-harry-clinton-global-initiative-601a35a7f8d1f424271069011730911d)

• Office of Governor Gavin Newsom: California signs child-safety chatbot and social-media laws (https://www.gov.ca.gov/2026/09/10/governor-newsom-signs-the-strongest-child-safety-chatbot-and-social-media-laws-in-the-nation/)

• California Legislative Information: SB 1119, Companion chatbots: children’s safety (https://leginfo.legislature.ca.gov/faces/billNavClient.xhtml?bill_id=202520260SB1119)

• NPR: 5 tips for parents worried about their kids’ screen time and online safety (https://www.npr.org/2026/09/28/nx-s1-5965849/screens-teens-parents-social-media-safety)

Factual update note: This explainer reflects the California law and the four listed reports available on October 7, 2026. SB 1119’s main child-safety chapter is scheduled to become operative July 1, 2027; implementation details and legal challenges may change how the rules work in practice.