California mandates crisis safeguards and audits for kids' chatbots

What this is about

California Governor Gavin Newsom signed a package of online child-safety laws on 10 September 2026. The measure most relevant to AI providers is SB 1119, known as Adam's Law. It tightens the rules for companion chatbots: systems designed not only to answer questions but also to simulate closeness, friendship, or emotional support.

The law requires crisis protocols when a child expresses suicidal thoughts, parental controls, and notifications when a child disables safety settings. Providers must also conduct independent child-safety audits and annual risk assessments. The rule is part of a wider package that also addresses addictive social-media features and the handling of student data.

What Adam's Law actually does

SB 1119 applies to operators of companion chatbots that may be used by minors. According to California's official summary, these services must translate safety promises into reviewable processes. That includes defined responses to statements about self-harm or suicide, tools for parents, and alerts when protective features are turned off.

The key difference from voluntary safety policies is external scrutiny. Companies must have their systems independently examined for risks to children and reassess those risks every year. Product design, moderation, escalation paths, and documented effectiveness therefore come under review. A warning label at the edge of an app is not enough.

The law does not prohibit companion chatbots, nor does it mandate one specific technical method. Providers retain room to choose how they comply, but they must be able to show that their measures exist and are regularly evaluated.

Why it matters

Companion chatbots can be especially persuasive to children because they are always available, remember personal details, and adapt their language. That closeness can become dangerous when a system validates harmful statements, encourages dependency, or responds badly during a crisis. A minor may easily mistake a fluent response for informed and dependable advice.

Adam's Law therefore shifts part of the responsibility away from children and parents and back to the provider. The independent-review requirement is especially important because it can expose differences between a safety feature advertised to the public and its real-world behavior. California may again provide a template for other US states, although the law does not create a uniform national standard.

The measure also matters to European providers. Any company offering a service in California will need to assess the local requirements. More broadly, it shows the kind of evidence regulators may increasingly expect: documented risks, testable safeguards, and clear escalation paths rather than general safety claims.

In plain language

A companion chatbot is a little like an always-available babysitter who speaks very convincingly but does not truly understand a situation. Adam's Law requires more than a house rule for that babysitter. It also calls for an emergency plan, controls for parents, and an independent person who regularly checks whether the rules work in practice.

A practical example

A 14-year-old tells a chatbot late at night that she sees no way out. Under the new approach, the service cannot treat that message as ordinary role-play. A defined crisis protocol must activate: the conversation should respond safely, display appropriate help, and trigger the escalation steps built into the product.

If the teenager then switches off a safety setting, the provider must, according to the official description, enable a notification to her parents. During the annual audit, reviewers would not merely check whether the relevant buttons exist. The important question is whether detection, response, and notification work reliably together under realistic conditions.

This example illustrates the rule's purpose, not every technical detail. The exact thresholds, deadlines, and review procedures will depend on the final statutory text and its implementation.

Scope and limits

• An audit cannot prove complete safety. Rare phrasing, ambiguity, irony, and language switching may still defeat crisis detection.
• Parental controls help only when they are understandable and difficult to bypass. They replace neither personal conversations nor professional care.
• California law is not a global standard. Jurisdiction, effective dates, enforcement, and possible litigation may change its practical impact.

There is also a privacy trade-off. A system may detect risks more effectively when it analyzes conversations comprehensively, but that analysis involves highly sensitive data from minors. Providers must therefore reconcile protection, data minimization, and demonstrable effectiveness. The information published so far does not show how well individual companies will meet those requirements.

SEO & GEO keywords

Adam's Law, SB 1119, California, companion chatbots, child safety, AI safety, crisis protocols, parental controls, independent audits, Gavin Newsom, children and AI