An overwhelming majority of likely voters want powerful AI systems to undergo mandatory formal security reviews before they are released to the public. That’s according to a new survey of Americans’ views on AI, which goes beyond the Trump administration’s existing policy of opt-in reviews for new advanced models.
For a limited time: Save 25% on NBC News subscription
Get exclusive reporting, live Q&As and ad-free reading.
The poll, conducted by the Washington, D.C.-based AI Policy Institute (AIPI), a nonpartisan research organization, found that Republicans were more enthusiastic about government security testing of AI models than Democrats, although more than half of voters supported such a measure regardless of their political affiliation.
The poll results are the latest sign of Americans’ bipartisan desire for stricter AI regulations as AI capabilities advance, and mark a departure from previous findings that Republicans were more skeptical of government intervention on AI issues than Democrats.
“We are currently seeing the government take a very active interest in managing the risks of AI systems and deciding which AI systems are safe enough to release,” said Peter Wildeford, director of policy at the AI Policy Network, a policy advocacy group affiliated with AIPI. “Americans want to do more for AI security.”
AIPI asked 1,007 likely voters across the country to choose between a small selection of answer options for each question. The survey was conducted on June 10 and 11 and required participants to sign up for the project through an online research marketplace.
The AIPI survey found that participants did not want to ban AI systems, although sufficient regulation would be possible. Given the choice between banning AI systems or requiring AI companies to implement security measures for their most advanced systems, two-thirds of survey respondents said they would prefer AI systems with guardrails.
However, when asked whether they would prefer to have AI systems without regulation or to ban AI completely, voters were strongly in favor of banning AI completely.
Government oversight of AI systems has emerged as a critical political issue over the past year. In early June, President Donald Trump signed an executive order targeting the cybersecurity capabilities of advanced AI models. The order directed federal agencies to strengthen federal cyber defenses and establish a mechanism to test new AI models for security concerns.
This mechanism has yet to be formalized or announced, and the order stated that any review would be voluntary for AI companies. Over the past month, the administration has been battling with Anthropic and OpenAI over the release of their latest models.
On Friday, OpenAI said it was forced to make its latest model, GPT-5.6, available to a limited group of trusted partners (rather than the general public) due to government requests over security concerns. The government on Friday gave Anthropic approval to give a number of trusted partners access to its most powerful Mythos 5 model.
The AIPI survey found that participants also prioritized regulatory oversight of data centers over outright bans on data centers. 47 percent of survey respondents said they would allow data centers if the AI systems being developed had security requirements and security standards, while 38 percent said they would ban data centers entirely. The rest of the respondents said they were unsure.
The proliferation of data centers has become a hot political issue across the country, increasingly serving as a proxy for Americans’ broader fears of AI systems. According to The Information, a tech news website, the data center boom in the U.S. is now facing more than 300 bans and moratoriums. Independent researchers recently found that data center opponents have blocked or delayed nearly $130 billion in projects this year.
In the AIPI survey, researchers found that over 60% of both Republican and Democratic respondents believed that the federal government – rather than AI companies – should set clear security standards for AI systems and then assess AI companies’ compliance with those rules. Most current security guardrails for AI systems are designed and implemented by AI companies.
Over 80% of respondents – 84% of Democratic participants and 83% of Republican participants – believed that AI companies should not develop AI systems that are smarter than humans until the companies can demonstrate that they can control the systems.
Federal efforts to regulate AI have faltered in recent months as advocates for stricter regulation clashed with the Trump administration’s view that such laws could hamper America’s AI industry and stifle innovation.
Just over two weeks ago, Commerce Secretary Howard Lutnick ordered leading AI company Anthropic to take down its two most powerful AI systems for national security reasons. Lutnick, along with other senior White House officials including National Cyber Director Sean Cairncross and Treasury Secretary Scott Bessent, feared that bad actors could use the systems to launch powerful cyberattacks.
The poll comes just days after several federal primaries devolved into a proxy fight over AI regulation. The race to replace outgoing Rep. Jerry Nadler, D-N.Y., in New York’s 12th Congressional District raised more than $40 million, most of it from political groups that advocate for AI issues.
Almost three-quarters of all respondents to the AIPI survey believed that AI will become a more important policy issue in the future.
The Pew Research Center released a survey this month showing that about two-thirds of Americans think AI is advancing too quickly.
Like the new AIPI survey, the Pew survey found that Republicans now have more confidence in the government’s ability to effectively regulate AI, a reversal from previous years.
In a 2024 Pew study, 70% of voters who identified as Republicans or leaned toward supporting Republicans said they did not have much confidence in the government’s ability to regulate AI, compared to 54% of Democratic-leaning voters.
However, in this month’s survey the dynamic had reversed. 74 percent of Democrats said they were not confident in the government’s ability to regulate AI, compared to 61 percent of Republicans.
“I think that people in the White House who have tried to have a no-rules-at-all perspective are out of sync with the American people with the results of the AIPI survey,” said Wildeford of the AI Policy Network.
https://www.nbcnews.com/tech/tech-news/voters-both-parties-want-tighter-ai-regulation-poll-finds-rcna351815
