The case for a comprehensive federal law to oversee AI

Congress considered 158 bills that mention artificial intelligence over the past two years, according to a count by the Brennan Center for Justice. But zero comprehensive AI laws have been passed.

There has been movement by states, however. In Tennessee, for example, the ELVIS Act, which protects voices and likenesses from unauthorized use by AI, became law in March. In Colorado, a law that takes effect in 2026 requires developers of high-risk AI systems to protect consumers from algorithm-based discrimination.

But some who fund AI technology say a federal law is needed. That includes Matt Perault, head of AI policy at the venture capital firm Andreessen Horowitz. The following is an edited transcript of his conversation with Marketplace’s Stephanie Hughes.

Matt Perault: We think it’s really important at the current moment to have a national competitiveness strategy for AI policy. We’re at a Sputnik moment right now with the release of DeepSeek, and we’re not going to be able to do that if we have a state-by-state patchwork where companies have to comply with different laws in different states. Our focus at Andreessen Horowitz is on Little Tech companies, so small companies and startups that are at the cutting edge of innovation and we think are going to help define the AI products of the future. For those companies, it’s particularly challenging for them to try to figure out, how do I comply with the law in California that might conflict with the law in Texas that might be different from the law in Florida?

Stephanie Hughes: Say more about that. How would navigating a patchwork of state laws even work when it comes to AI?

Perault: That’s the question — it doesn’t function particularly well. Like, if you think about other tech products that you use, like a messaging service, for instance. If you’re located in Baltimore and I’m located in Durham, North Carolina, and we’re sending communications back and forth, we want that to function seamlessly regardless of where we live. We don’t want to have a different experience in Maryland than the one that we have in North Carolina. And it’s the same thing for AI products. We are used to having tech products that function across borders. That’s actually one of the value propositions of many technology products, that it increases and improves communication and allows us to communicate more cheaply and easily. And having a patchwork where companies need to figure out, what does it mean when I build an AI product in one state versus in another, would be really challenging.

Hughes: AI laws can go in a lot of different directions. What do you think a federal statute — what should be the substance of it at this point?

Perault: There are a lot of different forms that it could take, but the focus for us is on ensuring that the regulation of AI technology focuses on harmful misuses, not on regulating AI development itself. If you focus on regulating harmful uses, you can do that through enforcement of consumer protection law, through enforcement of civil rights law, through enforcement of antitrust law. When you instead put the focus on regulating AI model development, really what you’re doing is creating a tax on math. You’re making it harder to do the innovation that actually would enable American companies to compete.

Hughes: So at Andreessen Horowitz, you guys think a lot about startups. You fund them, you benefit if they win. But why is it important for these startups to have clear regulation at a national level for them to really play in the space?

Perault: Startups have historically been the driver of innovation in our country. And our concern is that if you focus on regulating AI models, that that will literally make the process of startups engaging in the task of innovation more difficult. It functions essentially as a tax on them.

Hughes: How much unity is there within the AI community — big, established companies; tiny, little startups — about what regulation should look like, to your knowledge?

Perault: Well, a lot of the companies that we work with are incredibly small. They might not even have a general counsel. And so for lots of companies, they might not have a very specific vision of exactly all the different things that they want AI policy to look like. But they know that in order to compete, it’s going to be very challenging for them, and they don’t want to have to build really complicated processes to navigate complicated compliance burdens to try to identify what is happening in one state versus another. They want to ensure that they can build to a framework that allows them to innovate. They wouldn’t ask for an exemption from existing law. There’s no AI company that I’ve talked to that says, we want to be able to violate consumer protection law or we want to be able to violate criminal law. Obviously, there’s an understanding that you need to comply with the law. But the challenge of AI policy that focuses on model development is it actually goes into that garage and it makes it harder for the technical people to build the technologies that we think will allow American companies to compete with companies abroad.

Hughes: To talk about that for a second, everybody is paying a lot of attention right now to the abilities of the Chinese startup DeepSeek. How does regulation in the U.S., or a lack of it, affect how [China] can compete on the international stage, how its companies can compete internationally?

Perault: So I think this is an important moment to reorient American AI policy. If you focus on model development, you literally slow the process of developing models. And so now we’ve seen a model that can compete aggressively with American products. If you want to ensure that American products can keep pace and that the products that we use in the future are American products, as opposed to Chinese products, for instance, then if you are simply regulating model development, you’re putting American companies at a disadvantage. So from our point of view, that approach to policy is not the optimal framework for bolstering American competitiveness. I think the question then is, well, are you saying that we should take no action? And that’s not the position that our firm takes. Our firm is advocating for policy that really can protect consumers. But you would do that by addressing potential harmful uses of AI technology by consumers, not by aiming for the bank shot of hoping that if you slow model development, then you’ll also have potentially have the effect of reducing consumer harm, because the thing that gets caught up in that is really beneficial innovation.

Hughes: It’s 2025, we have a new Congress and a new administration. Mark Andreessen, a founding partner in your firm, is said to be advising President [Donald] Trump. How do you think these new policymakers could affect the likelihood of a federal AI law coming to fruition?

Perault: Well, we’ll see. But I think that there is a moment now, not just with the change in administrations, but also with the news of DeepSeek, to think about what the AI policy is that we want. Do we want to tax AI model development, or do we want to focus on actually enforcing existing law and looking at, to the extent that we need new law, focusing that on consumer protection itself?