AI chatbot child safety moved from voluntary product design into California law on 10 September 2026, when Governor Gavin Newsom signed Adam’s Law, SB 1119. The measure adds specific protections for children using companion chatbots, including crisis protocols for suicidal ideation, parental controls, notifications when a child disables safety settings, independent child-safety audits and annual risk assessments. The governor’s announcement describes the package as part of a broader set of child-safety rules for AI chatbots and social platforms.
The significance is not that California has discovered whether companion chatbots are safe or unsafe for children. It is that companies serving children now face a clearer legal expectation to identify risks before those products become deeply embedded in a young user’s routine. The law is aimed at systems that can behave less like a search box and more like a persistent conversational relationship.
What California’s AI Chatbot Child Safety Law Requires
Adam’s Law is designed around practical controls rather than a general instruction to “be safe”. According to the state’s announcement, companion chatbot providers must maintain protocols for suicidal ideation, provide parents with controls, alert parents when children switch off safety settings and undergo independent child-safety audits and annual risk assessments. The office of State Senator Steve Padilla, who authored SB 1119, describes the law as a comprehensive child-protection framework for companion chatbots.
Those requirements matter because conversational systems can hold attention for long periods and respond in a personalised tone. A child may treat a chatbot as a tutor, entertainer, confidant or friend without separating those roles as neatly as an adult product designer might. LiveAIWire has examined the developmental questions around children growing up with AI, where the unresolved issue is not simply screen time but what kind of interaction is replacing or supplementing human conversation.
California had already begun regulating companion chatbots before this new law. The 2026 action is therefore best understood as a tightening and expansion of protections, not the state’s first encounter with the category. The new event is Adam’s Law and its child-focused obligations, particularly the auditing, parental-control and risk-assessment requirements.
Why Companion Chatbots Create a Different Kind of Product Risk
A conventional app can be harmful through its content, design or data practices. A companion chatbot adds another layer because it responds directly to the user and can sustain a conversational relationship. That makes safety highly dependent on how the system reacts in difficult moments, how much it remembers, how it frames itself and whether a child understands that the apparently attentive personality is software.
The market has already moved beyond classroom-style question answering. LiveAIWire previously covered the AI companion market for children, including products built around ongoing interaction rather than one-off information retrieval. That kind of relationship is exactly why crisis handling and parental visibility matter. A system designed to keep a conversation going may encounter distress, dependency or sensitive disclosures that were not central to traditional children’s software.
What This Means for Parents and Product Makers
For parents in California, the law should create more visible controls and more formal accountability behind the product. It does not remove the need to understand what a child is using. A compliant audit cannot tell a family whether a particular chatbot fits a child’s maturity, temperament or reason for using it. The practical questions remain familiar: what does the system remember, when does it escalate a crisis, can a parent see or change safety settings, and what happens when a child asks for advice that would normally require a trusted adult?
For chatbot companies, the change is more structural. Risk assessment and independent auditing force safety claims to become inspectable processes rather than marketing language. A provider may need to show how its product responds to foreseeable child-safety scenarios, how settings work in practice and how risks are tracked as the model or product changes.
That may also alter product incentives. A feature that increases engagement but makes parental controls easier to bypass can no longer be judged only by retention or growth. Once the law places a formal duty around safety settings and risk review, the design team has to account for a second audience: auditors and regulators asking whether the product’s safeguards work as described.
The Law Does Not Prove the Safeguards Will Work
The most important limitation is timing. These rules have just been enacted. Their practical effect will depend on implementation, enforcement, audit quality and how companies change their products. An annual risk assessment can expose a problem, but it cannot guarantee that every harmful conversation will be prevented. Parental controls can create visibility, but only if families understand and use them.
There is also a wider jurisdictional question. California can shape product design well beyond its borders because large technology companies often prefer one consistent system to multiple state versions. That influence is possible, but it is not automatic. Companies could apply some protections nationally, restrict products by age or region, or challenge parts of the framework. The new law establishes obligations in California; it does not by itself create a national standard.
A New Baseline for Child-Facing AI
What has changed is the starting point. Companion chatbot safety for children is becoming something companies are expected to demonstrate, document and audit. The law recognises that a conversational AI used by a child is not merely another content service. Its behaviour during moments of vulnerability can matter as much as the information it supplies.
That is a forward-looking shift. As chatbots become more persistent, expressive and personalised, governments will increasingly regulate the relationship around the model, not only the model itself. California’s new rules provide an early test of whether formal audits, parental controls and crisis protocols can keep pace with products designed to feel increasingly human.
About the Author
Stuart Kerr is Technology Correspondent at LiveAIWire, covering artificial intelligence, cybersecurity and the social impact of emerging technology. LiveAIWire is an independent, human-led technology publication using AI-assisted research, editorial production and original AI-assisted editorial illustrations under his direction.