California's new AI chatbot law, signed September 10, targets "companion" bots, not the customer-service and lead-gen chatbots most small businesses run. If your bot answers product questions, books appointments, or qualifies leads, this changes nothing for you. If it has a name, a memory, and a personality built to feel like a friend, it's worth two minutes to check which side of the line you're on.
Here's what actually changed, who's covered, and what I'd check this week if you run any kind of AI chat on your site or WhatsApp number.
What Does California's New AI Chatbot Law Require?#
Governor Newsom signed SB 1119, known as Adam's Law on September 10, 2026, alongside a dozen other child-safety bills. It builds on SB 243, California's existing companion-chatbot law, in effect since January 1, 2026.
Adam's Law adds independent child-safety audits and annual risk assessments reported to the state attorney general. For any user known to be a minor, it sets hard defaults:
- Ephemeral chat mode, no saved history by default
- No push notifications between midnight and 6am, or during school hours
- A one-hour conversation limit
- A two-hour daily cap
Most of these duties phase in starting July 1, 2027. The law also bars a chatbot from implying it's human or conscious, or framing spending money as the price of keeping the relationship going.
Does California's Chatbot Law Apply to Your Business Bot?#
Almost certainly not. SB 243's definition of "companion chatbot" is functional, not technical: it covers systems built to sustain an ongoing, emotionally adaptive relationship. Customer service, business operations, and productivity bots are explicitly carved out.
Where it gets less clear is a chatbot with a strong persona that's meant to feel like a friend, not a form. We built exactly that kind of bot for a premium apparel client, a Gen-Z-toned Voiceflow concierge that carries a product carousel and answers style questions in character. It's still a customer-service bot by function, so it sits outside SB 243. I'll admit our first instinct was that any named bot with memory needed a second look. It doesn't, as long as the job is still support or sales, not companionship.
The Honest Trade-Offs#
The upside: this is a narrowly scoped law. It doesn't touch general-purpose assistants, voice devices, or game characters, and it leaves the SB 243 customer-service exemption intact rather than rewriting it. Read the whole thing, change nothing, and move on, for most of you.
The downside: enforcement isn't limited to the attorney general. SB 1119 opens a private right of action, so a parent can sue directly instead of waiting on a state investigation. My read is that this makes the exemption boundary something to document, not just assume. Nobody's tested in court yet exactly how far "emotionally adaptive" stretches, and that ambiguity is exactly what produces an expensive first test case.
How We're Handling It#
We reviewed Adam's Law and the underlying SB 243 exemption text the week it passed, because a chunk of our chatbot builds for D2C and apparel clients lean on persona and tone by design. Our rule now: keep the bot's job description in writing, support, sales, or booking, never "companion," and scope memory to the current session and order history, not an open-ended relationship the bot exists to sustain.
We build WhatsApp and web bots the same way for clients across the GCC, Europe, the USA, Canada, and Indian SMBs, regardless of which state's law is in the news that week. It's the safer default, and it reads clearly in an audit either way. We're not rewriting any live client bot over this. None of them are close to the companion line. It's a line item we check before we ship a bot with a persona now, not after.
What I'd Tell a Client Asking About This#
If your bot handles support, sales, or bookings: nothing to do this week. If your bot has a name and a personality and remembers past conversations to build rapport, get one sentence in writing describing its purpose as functional, not relational, and don't market it as a friend, especially anywhere a minor might reach it. Ten minutes, not a rebuild.
Still not sure which one you are? That uncertainty is usually the answer.
We've written before about what a business-grade WhatsApp bot actually looks like and about Meta folding Stilla into its own Business Agent this same week (worth a read if a persona-driven bot is on your roadmap). And if a consumer AI agent messaging through WhatsApp changes your risk picture, Meta's Muse launch is the other side of that story.
If you're building anything with a personality on the other end of the chat, send me what you're running and I'll tell you straight which side of this you're on. [cal.com/webepex/growth-review]
