Senators on both sides of the aisle are racing to put forward proposals for federal regulation of artificial intelligence, but most of the attention is on Senate Commerce Chair Ted Cruz, whose broad plan is popular with Big Tech but has yet to win over privacy and consumer advocates. The Texas Republican put forward a legislative framework last month, calling for a “light-touch” approach, though he unveiled only one bill, which would let AI deployers and developers request long-term waivers of federal regulations. It's not the only way he hopes to advance AI development. Cruz fell short on getting a moratorium on state AI regulations included in the recent Republican budget reconciliation law, but said recently he is not giving up. Meanwhile, other AI-related proposals have emerged: a "road map" from Sen. Mark Kelly, D-Ariz., that would tax AI companies to address social and environmental harms, and a bipartisan bill from Sens. Richard J. Durbin, D-Ill., and Josh Hawley, R-Mo., that would change how AI companies are held accountable for harms. Still, Cruz might stand the best chance of making good on his agenda, given his position atop the Senate Commerce Committee and potential support for his plan from the giant tech firms that are pushing AI throughout their products. Playing in the ‘sandbox’ That framework, which Cruz said was based on “five pillars,” has mostly remained a policy proposal, but included his bill that would establish a federal regulatory “sandbox” for AI. The bill would allow companies to apply to waive federal regulations relevant to their AI systems for up to 10 years, with the White House — through the Office of Science and Technology Policy — designated to oversee the program. The proposal is already attracting criticism from groups concerned with the potential misuse of AI, and those that fear companies would be given leeway that violates civil freedoms and consumer protections. Cruz sought to address those issues when he issued the framework, saying a sandbox was not a “free pass.” “People creating or using AI still have to follow the same laws as everyone else,” he said. “Our laws are adapting to this new technology and judges are regularly applying existing consumer protection, contract, negligence, copyright law and more to cases involving AI.” The bill would require companies to explain how waiving regulations would benefit customers, efficiency, or innovation. It would also require developers to list foreseeable risks and explain how they’re outweighed by benefits. The bill would require agencies approving applications to identify how consumers would be protected. Further, the White House would annually ask Congress to amend or repeal regulations without which the program has shown that AI companies are "able to operate safely." The sandbox bill is not universally popular with the GOP. Hawley said he doesn’t like the idea, and that the level of regulation is already low. “The companies are going to be fine. Like it's . . . a myth that the companies need this, that or the other thing from the government,” he said. “The companies are currently doing whatever the heck they want. They're the richest companies in the world.” Broader approach Regulatory sandboxes, even those focused on AI, aren't new. Texas and Utah have passed laws to establish AI sandboxes and Delaware is planning its own program. However, Cruz’s proposal would last longer and likely include more companies. That’s according to Sophie Tomlinson, the director of programs for the Datasphere Initiative, a Switzerland-based research and policy group focused on data governance. Tomlinson described sandboxes as “safe spaces” to test tech and practices while supporting innovation and “regulatory agility and flexibility.” In Tomlinson’s experience, most sandboxes run for six months to a year, sometimes extended by the presence of a new cohort after tweaks have been made to the program after a first round. Sandboxes, Tomlinson said, can have different goals. Some are more focused on compliance, while others are about creating “a regulatory kind of leeway, in a controlled environment for a specific period of time.” She emphasized the importance of clear governance around accountability and potential risks, including rules about sharing data or intellectual property within a program. She also said it’s best to be specific about what criteria a sandbox is meant to address. In the case of a U.S. sandbox, she noted that the program wouldn’t be capped at a certain number of companies participating. “A broad approach is quite interesting, but at the same time it needs to be . . . meaningful. There needs to be the oversight and the risk mitigation too, and also for the regulators to actually have useful learning so they can adapt too, right, to the innovation.” An industry lobbyist who spoke on condition of anonymity called Cruz’s bill “a great way to address some of the regulatory challenges that specifically the AI deployers face,” but said that AI developers are more concerned with state-level regulations. The lobbyist also worried that the White House science office won’t have the necessary authority to implement the program. The lobbyist said the legislation should give power to individual agencies for sector-specific sandboxes. Amy Bos, director of state and federal affairs for industry group NetChoice, which counts Amazon.com Inc. and Google LLC as members and works for “free enterprise and free expression” online, sees the bill as a good start toward letting “innovators innovate.” “It solves, really, a fundamental problem, and that is . . . our current regulations were written for a pre-AI world,” Bos said. “So . . . right now, I think you're seeing AI applications, they can be blocked by decades-old rules that simply just don't make sense right now for modern technology. And so the sandbox bill gives companies . . . these temporary waivers while maintaining . . . explicit safety protections against health, against public safety risks.” The sandbox proposal is less popular with those concerned about the risks AI could pose. The bill would require companies to notify the sandbox program of any incidents that cause harm, including to health and safety or to the economy. The bill would define health and safety risks as including bodily harm, death, or substantial adverse health effects. It would define economic risk as likely “tangible, physical harm to the property or assets of a consumer.” Cody Venzke, a senior policy counsel for the American Civil Liberties Union, said the bill would give AI developers and deployers “roving permission” to violate important rules for consumer protection and civil rights. “Sandboxes, I think, should not be viewed as some sort of moderate compromise,” Venzke said. “They are essentially giving entities permission to be exempted from existing law. And these are laws that apply to everybody else. And I don't see a good reason why we should say to artificial intelligence, ‘You are special. You get out of jail for free.’” Moratorium redux? Cruz’s bill doesn't include a federal preemption of state laws or a moratorium on new regulations, but such a moratorium could still be coming. He said at an event last month that “I still think we'll get there and I'm working closely with the White House,” but that there’s not yet a public-ready plan for how the moratorium could move to the floor. An amended proposal by Cruz was removed this summer from the reconciliation bill on a 99-1 Senate vote. The industry lobbyist said advocates are looking at “all available options.” “I’ve been in conversations that are looking at every moving piece of legislation before the end of the year,” the lobbyist said. Bos said companies want to avoid a “nightmare scenario” of navigating different states’ regulations on AI. “We’re hopeful that legislation can move yet this year. It's certainly, we've heard, a priority for the Hill,” she said. “I think it has backing from the White House, from the administration.” Hawley opposes a moratorium, calling it “crazy” to preempt states that have legislated to protect children from online exploitation or deepfake porn. “I want to protect kids from all those things, and I want the states to be able to do that. You know? I mean, I think it's a terrible idea to say that, ‘Oh no, states, you can't protect your kids in your own state,’” Hawley said. Venzke opposes preemption because he doesn't want to see states' role "displaced" without federal regulations in place. “States are, you know, famously, the laboratories of democracy, and they are already working on addressing these issues, and that will help the federal government formulate what its response is.” A 'light-touch' framework Outside of the sandbox bill, Cruz’s framework promises multiple goals: streamline AI infrastructure permitting, open federal data for model training, “protect free speech,” “prevent nefarious uses of AI”, and “defend human value and dignity.” Bos said “the devil is always in the details, with the other bills, but we like where this is headed.” The industry lobbyist said it is a "little bit early" to know how the framework will be translated into actual bills. “I’d be really interested to see how he intends to write legislation protecting free speech in the age of AI and trying to defend human value.” The lobbyist also worried that the free-speech portions of Cruz’s plan, especially if it goes down the road of addressing “woke AI” could lead to losing the possibility of a bipartisan path. “Those kind of ideas are more about messaging than actually trying to work with allies in Congress to advance some of these principles,” the lobbyist said. Hawley said he takes a “somewhat different view” from Cruz on AI, and is particularly worried about workers and data use. “I just think that we've got to put guardrails in place that protect people and guarantee their rights,” he said. In an election year, the lobbyist said, even bipartisan areas of AI policy could be politicized. “In terms of like, bills passed and things actually moving through Congress . . . I’m pretty pessimistic,” the lobbyist said.
