What are AI companion chatbot laws, and do they reach schools?

Twelve states enacted AI companion-chatbot laws during 2026 — California, Colorado, Georgia, Iowa, New York, Rhode Island, Idaho, Nebraska, Oregon, Connecticut, Washington and one further state, with Hawaii's bill reaching the governor and Arizona's vetoed. They regulate a specific product category, not AI generally, and most educational tools sit outside the definition. Knowing where the line is drawn is useful whether you are choosing software or explaining risk to a teenager.

The definition, and why it is narrow on purpose

The statutes converge on a system "designed to simulate an ongoing human-like relationship," one that remembers past interactions and holds "continuing discussions about a user's personal life or emotional well-being."

Three elements have to be present: relationship simulation as a design intent, persistent memory of the person, and conversation about their inner life. A tool that answers questions and forgets you is not covered.

The exemptions confirm the intent. They typically carve out customer service and commerce tools, business productivity and research applications, internal enterprise systems, developer tools, virtual assistants, and AI features in video games.

What the laws require

Disclosure on a timer. Periodic reminders that the user is talking to a machine — commonly every three hours, and hourly for minors in some states. The interval matters: legislators concluded that a single notice at the start is not enough once a conversation runs long, which tells you what they were worried about.

Crisis protocols. Documented procedures for self-harm and suicidal disclosure, typically routing users to the 988 Suicide and Crisis Lifeline in the United States.

Minor protections. No sexual or romantic content with minors, no claims to be human or sentient, and — the provision worth dwelling on — prohibitions on design intended to foster emotional dependency.

That last item is the tell. It exists because attachment is a monetizable product outcome, and several legislatures decided that engineering it into a product used by children is not acceptable.

Do these laws reach schools?

Mostly not, and for a defensible reason. A tutoring assistant, a writing feedback tool or a research notebook is a productivity or research application — it does the task and stops. None of them are designed to be your child's friend.

Where schools should pay attention is at the edges. A "student wellbeing companion," a character-based tutor with a persistent persona and memory, or any product whose marketing promises a relationship rather than a result, is much closer to the line. The question to ask a vendor is not "are you compliant" but "does your product remember a student's personal life, and is it designed to be returned to for its own sake?"

What this means for a family

The distinction is more useful than the law. Companion products and assistant products are different things with different business models, and a teenager can learn to tell them apart in about two minutes.

An assistant wants you to finish and leave. A companion wants you to stay. Persistent memory of personal detail is usually the feature that moves a product from one category to the other, and it is often on by default and usually possible to turn off.

If a conversation with any of these products has touched on self-harm, that is a moment for a person — a GP, a school counselor, or a crisis line — rather than a settings change.

Why this matters for curriculum

An AI literacy course that only teaches "how to use the tools" has nothing to say about this, and this is where the actual risk to a fourteen-year-old lives. The teachable content is concrete: what a product is designed to optimize for, why a system that remembers you is different from one that does not, what it means that a machine has no beliefs or feelings to have, and who profits from your attention.

Idaho's approved standards put the same idea in a single line — students should analyze the difference between human and AI interaction "by recognizing that AI lacks sentience, emotions, and personal beliefs." Teaching that properly is a better protection than any filter.

A note on ZEROTH's own design

ZEROTH includes a source-grounded research assistant built over the course files. It is squarely inside the research and productivity exemption and we intend to keep it there: it does not remember a learner's personal life, does not ask about feelings, and is not written to be anyone's companion. That is a deliberate design line, and we check the course copy against it, because a curriculum about AI ethics that shipped a companion bot to teenagers would deserve everything it got.