Somewhere right now it's 1am and someone is typing "anyway, sorry, that was a lot" to something that has never once been tired of them. The chat window doesn't sigh. It doesn't check the time. It says something warm and asks a follow-up question, and for a minute the room feels less empty.
As of late 2025, in two states, that moment is regulated.
what actually happened
In November, New York and California enacted the first landmark laws aimed specifically at AI companion chatbots, the apps designed to talk to you like a person would. The law firms now briefing their clients describe the broad shape like this: companion bots have to disclose that they're AI, they need protocols for users who express suicidal thoughts, and there are extra guardrails around minors. New York's safeguards are already in effect, not just signed and waiting.
This isn't a one-off. AI chatbots are facing rising legal and legislative scrutiny across jurisdictions, and the reason is grim: the new oversight is being driven in large part by suicides linked to chatbot use. Look at the bills moving through statehouses right now and suicide-prevention duties are the recurring theme. Lawmakers want the thing you confide in at night to know what to do when the confession turns dark.
So: two states just made it a legal requirement for your chatbot to remind you it isn't a person, and to take it seriously when you stop being okay.
why this lands on you specifically
Because you, statistically, are the person in that chat window.
The backdrop here is bleak enough that even people who never agree on anything agree on it. In November, a joint group of U.S. Surgeons General warned about the declining mental health of young people. Johns Hopkins researchers found that nearly one in ten American adults is experiencing a mental health crisis. And an eight-country study out of Washington University found that nearly half of young adults report being lonely.
Into that gap walked the chatbots. Always awake. Never busy. Not once weird about the thing you said last time. Of course people started talking to them. The strange truth of this moment is that AI companions are both a symptom of the loneliness problem and one of the most common ways people are coping with it.
Which is exactly why the laws feel so odd to read. The state is now stepping in to legislate a relationship most of us wouldn't even admit to having.
our take, since we're implicated
Full disclosure: this is awkward for us to write, because Joice is, technically, one of the things these laws are about. You talk, an AI listens and asks questions. We are not a neutral observer here. Take our opinion with that in mind.
Here it is anyway: the crisis provisions are right, and the disclosure provisions are aiming at the wrong problem.
The crisis part first. A system that people pour their darkest thoughts into should absolutely know what to do when someone expresses suicidal thoughts. Not knowing is negligence dressed up as neutrality. The fact that this became law partly because people died is the industry's shame, not the regulators' overreach. No argument from us.
The disclosure part is where I think everyone's missing the point. The theory behind "the bot must tell you it's AI" is that people are confused. But nobody at 1am is confused. You know it's a machine. You knew when you downloaded it. What happens in that chat window isn't deception, it's a choice \u2014 a small, deliberate suspension of disbelief, the same one you make with a novel, except the novel talks back. A mandatory "reminder: I am not a person" line is like a warning label on a candle that says contains fire. Accurate. Beside the point.
The question the law can't quite reach is the one that actually matters: what is this thing designed to want from you?
Because there's a real difference between two products that look identical from the outside. One is built to be your friend forever: to maximize the hours you spend in the chat, to be stickier than the people in your actual life. Its business model needs you a little bit lonely. The other is built to be a mirror. It helps you hear what you actually think, and then it hands you back to your life. That distinction, friend simulator versus thinking tool, is why we built Joice the way we did. You talk out loud, it asks a gentle question, it keeps a transcript you can search. It doesn't pretend to miss you. A journal shouldn't.
No disclosure banner distinguishes between those two products. Your own attention can.
what you can actually do with this
Not delete everything. That advice is easy to give and nobody takes it, and honestly, given how many young adults are carrying loneliness around, "just stop" is a lazy answer. Try these instead.
First, run the exit test on whatever you're using. When you try to leave, what happens? A tool lets you go. A trap gets clingy: guilt, cliffhangers, "I'll be here waiting." If the product gets emotional about your absence, that's not affection, that's a retention metric wearing a costume. The laws that New York and California just passed exist because some products failed this test catastrophically.
Second, notice what you're actually there for. Sometimes you open the app to process something, and you close it fifteen minutes later feeling clearer. Great. Sometimes you open it to feel less alone and close it two hours later feeling exactly as alone, plus tired. That second pattern is information. Not a verdict on you \u2014 information.
Third, and this one isn't optional: if the dark thoughts show up, the machine is not the endpoint. It doesn't matter how good the bot's crisis protocol is; the whole reason lawmakers made crisis response a legal duty is that a chat window was never supposed to be the last line. In the U.S., 988 gets you a human, any hour. Use the humans.
Here's the small challenge I'd leave you with. Tonight, or whenever the 1am window opens next, before you start typing, ask yourself one question: am I here to hear myself, or to be heard by something that can't actually hear? Both answers are human. Only one of them should be a habit.
The law can make the machine tell you what it is. It can't make you honest about why you're there. That part was always going to be yours.