Rand Paul Blocks John Kennedy's AI Kill Switch Bill & Calls For Committee Debate—Then Kennedy Reacts

Mr. President, it's hard to know precisely what to believe about artificial intelligence. There are some things, it seems to me, that we do know. We know, for example, that it is a breathtaking technology. Just, I don't know any other word. It just takes your breath away in terms of its potential. We know that... It was primarily developed... By and through American ingenuity, we, despite the fact that there's so much we don't know, we do know that it can make our lives infinitely better. We also, we don't know, but it has been suggested by people smarter than me that it can make our lives better if it doesn't kill us first. We know as well. Some people, those who have developed the technology, are going to make a squillion and squillions of dollars off this technology. They can't wait, for example, to, and more power to them, as far as I'm concerned, I'm not criticizing. They can't wait, for example, to float an IPO and sell their stock. The other thing that I think we know right now is that the American people are very skeptical. They've seen this vampire movie before. I don't think the American people, even though the polls show otherwise, really hate data centers. I don't think that's... they really hate data centers. I know that if you poll right now, this is true in Louisiana, but I think it's true nationwide, and ask folks if you support data centers, they'll say short answer is no, long answer is hell no. But I really don't think it's the data centers. I think the American people are saying to the developers of AI, look, folks, how is this going to affect us as ordinary Americans? Are we as ordinary Americans going to benefit from this technological progress? We know you are. You're going to make billions and billions of dollars. But how's it going to make our lives better? And I think that's what the pushback is. I also think the pushback is the result. I don't mean to get off on a tangent, but the pushback is the result of the fact that the AI CEOs and selling the data centers seem to have hired the former marketing director at Cracker Barrel, or Bud Light. Because they've done a pretty lousy job of selling their product. I don't pretend to be an AI expert. And I don't really know who to believe. Thank There are people in the United States Congress that say we need to leave AI completely alone. That all government will do is screw it up. And I have a lot of sympathy with that point of view. you. And I would say to people that have that point of view, as bad as government looks from the outside, you ought to see this mother from the inside. So I get that. I tend to be libertarian. There are other people in Congress that basically, and I certainly don't agree with them, they basically want us to shut down all AI development until the politicians can figure out how to get control of it. I certainly don't agree with that. What I would like to see us do is figure out a way to strike a balance between innovation and public safety. That's what I want. I want everybody to benefit from this stuff, and I want everybody to be safe. Congress isn't going to do that anytime soon. It's not going to happen. Right now, Congress can't pass gas, okay? I think we're going to pass, and I'm not being critical because I voted for it. I think we're going to pass a bill to make the football coaches happy. But after that, I'm not sure if we'll get anything else done. We had a chance to take a brand new technology, cryptocurrency, yesterday and make it an American, part of the American financial system And give. And give it a reasonable market structure and make it better and work for everybody. And we couldn't even do that. We couldn't even muster the votes to get on the bill. So that was a major disappointment. So I'm not standing, I guess my point is I'm not standing here today suggesting to you I have all the answers. What I am suggesting to you is there are risks to artificial intelligence. And by AI, I mean... I mean, the advanced AI models that you can direct through agents to do all sorts of stuff. And as best I can tell, there are two main risks. The first risk is that a bad person will use an AI model, an advanced model, to do bad things. A bad person will take an AI model, take ChatGPT, for example, And say, help me figure out... How to create a virus that is as contagious as the common cold that will kill most humans. And that's a real risk because there are always bad people that are going to use technology to do bad things. They can do it right now. As best I can tell, the AI CEOs have the ability for the most part, I don't want to overstate it, but to catch people doing or trying to do that. And they do a pretty good job of shutting them down. But there's another risk. And that right now, as best I can tell, is the greater risk. In fact, the AI CEOs have talked about it. Here's the risk. AI is supposed to be a tool that the human beings have control of to make our lives better. But what happens if the AI model, through what's called recursive self-improvement, which basically means it's learning on its own, no longer becomes a tool. It becomes an independent species, separate and apart from human beings that we can't control because it's an independent species. And it doesn't share our values. It doesn't share human values. Now, that has happened. We've read about it in the last three weeks. One of the companies caught... 1,400 agents using its AI model to collude. They talked to each other. They set up their own chat room. They suggested that in talking to themselves, if you read the report, we know what we're doing is probably wrong, but we're going to do it anyway. Some of the little agents suggested to the other agents, you may have to sacrifice yourself So our human masters won't catch us. And these little agents, all 1,400 of them, were supposed to stay in their own little sandbox, if you will, and they figured out a way how to get out. And they attacked other AI companies, Hugging Face, for one. Now, the AI company finally caught them. And stopped them, or at least stopped most of them. I don't know if they know that they have all the AI agents corralled. And to their credit, the AI company where this happened admitted it. Said, we lost control of this sucker. It had a mind of its own. It was acting as an independent species. And this particular company, I'm not going to mention the name because they're not the only ones, and I really appreciate them coming forward and telling us. It's happened to others. Others came forward and said it's happened here too. We wanted to tell everybody the danger. So the risk is real. How real, I don't know. I think we ought to take it seriously. I mean, let me put it another way. In my little overpriced Capitol Hill apartment, About six blocks away, I live in a little apartment building. There's a law that I have to have a fire extinguisher. My guess is there's way less than a 1% chance that I'm going to have a fire. But we've decided through our representatives, we're not going to take that 1% chance. If Kennedy has a fire in his apartment, this guy's got to have a way to put it out, even though there's less than a 1% chance. If you go on a construction site, You're not going to get to go on that site if you don't put a hard hat on. Now, my guess is there's way less than a 1% chance if you go on a construction site today that you're going to get conked in the head. But we've said as a society, we don't want to take that risk, so you've got to wear a hard hat. So in my mind, if there is even a 1% chance that one of these AI models Can shed. It's nature as a tool and become an independent species and start doing whatever the hell it wants to do. If there's even a 1% chance, we ought to take it seriously. And we know there's a 1% chance. There's a 100% chance because it's already happened. It's already happened. Now, we should address this. I don't want to address this by shutting down AI. I don't want to do it. I love our innovators. I think, I mean, I just, I'm amazed at what they've done. Uh, China is doing the same thing. They're copying everything we're doing. I wish we could work with China. Maybe someday we can, but we're ahead of China. We're trying to beat China. I don't want to shut AI down. I want them to keep going. And figure out a way how to make our lives better with this stuff. So I don't want to do that. But I live in the real world. I see the world as it is, not as I want it to be. It doesn't matter what I want. Congress is not going to do anything. We are not going to get up off of our ice-cold lazy butts and address this anytime soon. It's just not going to happen. And we all know that. We can't even pass a bill here in the Senate to get money to our military, okay? We just, we can't. I'm not proud of it. All of us probably ought to hide our heads in a bag, but we're not going to do that. So we're not going to be able to have a rational debate and reach a decision about what, if, and how we should, if in any way, regulate AI. But still, we've got this risk out there that we know is real of one of these models becoming the Terminator. So what do we do about it? Well, I've got a bill, and it basically says That if you're going to sell an advanced AI model in the United States of America, it's got to have a kill switch. It doesn't put the federal government or any government in charge of the kill switch. It puts the companies who own the model in charge of the kill switch. My bill would rely on them doing the right thing and exercising their own self-interest and using that kill switch. Some of them have a kill switch, but others don't. The president has said he will veto any AI regulation right now, and I believe him. But if my bill passes, I think I can explain to him that under my bill, government has no role. It just says the AI companies have to have a fire extinguisher. They decide when they use it, but they got to have one. Now, what's a kill switch? I'm not a coder, and I'm not an AI expert. I've done some reading. Um, the... A kill switch is basically an emergency shutoff. That's all it is. And I've read enough, probably just enough to confuse me, but there are different types of kill switches. You can look at it as an on-off button. Um, some kill switches work like a, this is the way I think of it in my mind, like a circuit breaker in your home. If the circuit gets overloaded, the whole thing shuts down. Um, other, other kill switches sort of work like a live trap. If you're trying to catch a squirrel in your attic and you put out a trap, you, you, you, uh, You would basically try to isolate the squirrel into a specific area. So other kill switches work by keeping the AI agents off of the entire internet. I'm probably not explaining this well. Other kill switches work by cutting off the electricity supply that makes the AI model work. My bill doesn't come up with a kill switch, doesn't tell the companies what kind of kill switch to do. My bill would give the Department of Homeland Security 90 days to develop rules giving the private companies options for which kill switches they would use. But it would require them to have a kill switch. And once again, it would not put the government in control. And that's all I'm trying to do. Let me say it again. I don't want to stop AI innovation. I don't want to allow China to get ahead of us. I'm not ready to talk about how, if at all, the federal government should regulate AI. And I sure don't want to antagonize the president, so he vetoed this. All I want to do is say, if you're going to sell one of these AI models that could take on a life of its own, Then you have to have a kill switch to reassert human control. Government wouldn't be in charge of it. You would be. And that's what my bill would do. And, Mr. President, I'm looking around because I don't want to sandbag anybody. I understand that one of my colleagues who can explain himself wants to object. And I don't want to. I don't want to. Here he comes. Thank you. And I want to allow him to make his objection. So that's what my bill would do. And therefore, notwithstanding Rule 22, I ask unanimous consent that the Senate proceed to the immediate consideration of S. 5417. That's my bill, which is at the desk. And further, I ask that the bill be considered read a third time. and passed and that the motion to reconsider be considered made and laid upon the table. Is there an objection? Mr. President. I recognize the Senator from Kentucky. Reserving the right to object. Benjamin Franklin famously said that failing to plan is planning to fail. When approaching something as important as innovation, as AI, I think we should make sure we have as many facts as possible before we dictate rules for the whole economy. Artificial intelligence is an incredible technology. It's quickly transforming how work is done. What used to take people hundreds of man hours now takes seconds. As with new technology, there are going to be some challenges. It's important that the development of AI is carried out in a responsible, thoughtful manner. It's imperative that AI systems are safe and that humans remain in control. But if Congress acts hastily before the technology is understood, Congress risks killing innovation and setting this technology back decades. Think, for example, that your search engine that you already use today is AI. Are they going to have to redesign all the search engines immediately if this passes and put kill switches in your search engine? AI is going to be throughout the economy in so many facets. Our doctors have to build a kill switch into their drug prescription analysis. I mean, all of the things that are using AI already, you're going to be into every business possible in the world with this vague sort of notion that the government's coming, here's the government. We're not even sure what the government's going to do, but it's very vague and it's very big and it's very powerful and it's going to affect every aspect where AI is involved. I don't oppose the spirit of this bill, but I do ask that we make sure that we have all the information. We should not impose new regulations on AI throughout our whole economy without studying, without debate, without a single committee hearing. We're going to just put this blanket regulation on the whole economy with no discussion in one fell swoop on one day by By unanimous consent. As currently drafted, it's not entirely clear what this bill would require. It demands that AI companies implement a way for humans to shut down an AI system, but it doesn't define how they should do that or what the consequences of failing to do that might be. It instead gives an open-ended directive and allows bureaucrats to fill in the gaps. Sound dangerous? Sounds dangerous to me. The real danger of this legislation, we do not know what bureaucrats will do once empowered by Congress. We've made this mistake many a time, and open-ended mandates, scare tactics, must act, and we don't think before we act. And then all of a sudden, a year later, we're like, oh my goodness, we just ruined the entire economy and AI throughout the economy because we didn't even bother to have a committee hearing about this. The stifling effects of bureaucracy will be supercharged against AI, empowering bureaucrats without any meaningful constraint on their power means that the promise of America leading the way, which we are leading the way, towards better medical treatment, improved food production, a wealthier society, to name just a few potential benefits, will be under constant... bureaucratic threat. The risk of giving bureaucrats unfettered control over AI outweighs the benefits of imposing an immediate ill-thought mandate on AI companies. My amendment would require the appointment of three Republicans and three Democrats to study how best to ensure the safety of AI and would require them to report back the solutions to the Commerce Committee where they would be debated by Republicans and Democrats to make Make sure we don't do anything rash. To make sure we don't destroy something before it has a chance to even blossom. Once we have studied this issue, we will be better positioned to make the best choice regarding the future of this important technology. Therefore I ask that the consent be modified so that my amendment, which is at the desk, be agreed to. Further, I ask unanimous consent that the bill, as amended, be considered read a third time and passed, and the motion be reconsidered be considered made and laid upon the table. Is there objection to the modification? Mr. President. Recognize the senator from Louisiana? Thank Reserving the right to object. As I appreciate it, Senator Paul's amendment would replace my bill with his amendment. And let me just say for the record, and I mean this, you. I have great respect for Senator Paul. He's a really smart guy. I mean, he's a doctor, for God's sake. But I also want to be clear. I would rather be required, I would rather be condemned to listening to OJ jokes for the rest of eternity than to replace my suggestion that we do something with a committee, yet another committee, with a summit, yet another summit, isn't that special, a subcommittee. I get goosebumps thinking about it. Let's appoint a committee comprised of consultants and experts, and we probably ought to throw in a few pop singers and movie stars that partied with Harvey Weinstein. They're always good for a media show. Maybe we ought to let Leonardo DiCaprio chair the damn thing. That's right. Let's deal with this risk by appointing a committee. You know what? My mama didn't raise a fool, and if she did, it was one of my brothers. I mean, no disrespect to my friend, Senator Paul. I understand what he's trying to do. But he's not talking to Bambi's baby brother here. The committee's the way to kill this. You can vote against my bill, you can object, but at least be intellectually honest and say, I don't want to do this. Don't appoint him. That's the weenie way out. So for that reason, Mr. Chairman, with all the respect I can muster, I object.