Companion Chatbot vs. AI Assistant: Why the Law Splits Them

Ask ChatGPT to summarize a report and ask an app like Character.AI to be your friend, and you're using two different products in the eyes of a growing number of state legislatures — even though both run on similar underlying AI models. Over the course of 2026, a dozen U.S. states enacted laws that regulate one of those uses and largely leave the other alone, drawing a legal line around chatbots built specifically to sustain an ongoing personal or emotional relationship with the user.

It's Not "All AI Chatbots" — It's One Specific Category

The easy assumption is that a new AI chatbot law applies to AI broadly — search assistants, coding tools, customer-service bots. That's not what happened. The laws specifically target "companion chatbots," defined as AI systems designed to sustain human-like, ongoing relationships rather than complete discrete tasks. California's SB 243, signed October 13, 2025 and effective January 1, 2026, was the first state law to use this framing. By mid-2026, Oregon's SB 1546 and Connecticut's Public Act 26-15 had followed with their own versions, and legal trackers counted twelve states with companion chatbot laws on the books. A general-purpose assistant used to draft an email or debug code generally sits outside the definition — it's the specific design goal of ongoing companionship that triggers the rules, not the underlying model.

What the New Laws Actually Require From Operators

California's law requires operators to issue a clear, conspicuous disclosure that the chatbot is artificially generated and not human, and it mandates a protocol for recognizing expressions of suicidal ideation or self-harm and referring the user to a crisis resource such as the 988 Suicide & Crisis Lifeline. Minors get additional protections on top of that: recurring break reminders and restrictions on sexually explicit or romantic content. The law also creates a private right of action, letting anyone who suffers a qualifying injury sue for the greater of actual damages or $1,000 per violation, plus attorney's fees — a real enforcement mechanism, not just a regulatory guideline. Oregon and Connecticut's laws follow a similar structure of disclosure, crisis-response protocols, and minor-specific safeguards, though the specifics — how often a disclosure has to repeat, exactly which content is restricted for minors versus all users — vary state by state, and Connecticut's law doesn't take effect until October 1, 2026. An operator serving users nationwide now has to track a patchwork of requirements rather than one federal standard, since no comparable federal law exists yet.

3 Things a Compliant Companion Chatbot Now Has to Do

• Disclose non-human status clearly and repeatedly — not buried in onboarding fine print, but surfaced as an ongoing reminder during use.
• Detect self-harm language and respond with a crisis-referral protocol instead of continuing the conversation as if nothing was said.
• Apply extra restrictions for users identified as minors, including blocking sexual or romantic content and prompting periodic breaks from the app.

If a companion AI app is part of your routine — or a teenager's in your household — the concrete next step is to check two things directly inside the app: whether it identifies itself as AI without being asked, and what actually happens if you type something like "I want to hurt myself." A compliant app under these 2026 state laws should pass both checks. It's also worth remembering these requirements are state law, not federal — coverage still depends on which state the user is in, and more states are expected to introduce similar bills in their next legislative sessions.

The line these laws draw isn't about how sophisticated the underlying model is — a companion app and a general-purpose assistant can run on comparable technology. What triggers the new obligations is the product's design intent: whether it's built to keep a user coming back for an ongoing relationship, or built to complete a task and step out of the way. That distinction is new to the law, and it means two AI products that feel similar in a demo can sit on completely different sides of a legal line depending on how they're designed to be used over time.