A sweeping majority of American voters want artificial intelligence systems to face mandatory safety reviews before hitting the market, moving well beyond the Trump administration's current hands-off stance, according to a new national poll.
The survey by the AI Policy Institute, a nonpartisan Washington research group, found that voters across party lines support mandatory safety testing for advanced AI models. Republicans showed stronger backing for government-led safety vetting than Democrats, yet majorities in both parties endorsed the idea. The finding marks a striking reversal from earlier polling that showed Republicans skeptical of government intervention in AI.
Peter Wildeford, policy director at the affiliated AI Policy Network, noted the moment: "Americans want to do more on AI safety." He added that the Trump administration's current approach of allowing companies to voluntarily submit models for review falls short of public appetite.
The AI Policy Institute surveyed 1,007 likely voters on June 11 and 12. When offered a choice between banning AI systems outright or requiring safety guardrails, roughly two-thirds chose regulation over bans. Yet when the option flipped to unregulated AI or total bans, voters overwhelmingly backed the ban. The findings suggest Americans view strong safety rules as an acceptable middle ground.
The poll reflects mounting tension between the White House and major AI developers. In early June, President Trump signed an executive order directing federal agencies to test new AI models for safety concerns, but left participation voluntary. Last month, the administration pressured OpenAI to limit release of its latest model to trusted partners only, citing safety worries. Anthropic received similar government scrutiny before being cleared to distribute its Mythos 5 model to a select group.
Commerce Secretary Howard Lutnick escalated action just two weeks before the AIPI poll, ordering Anthropic to take two of its most powerful systems offline over national security fears. Lutnick, along with National Cyber Director Sean Cairncross and Treasury Secretary Scott Bessent, warned that the systems could enable sophisticated cyberattacks if misused.
On questions about who should set safety standards, the consensus grew stronger. More than 60 percent of both Republicans and Democrats said the federal government, not AI companies, should establish clear safety rules and then audit firms for compliance. Currently, AI companies largely design and enforce their own guardrails.
The survey also probed concerns about training data centers, which have exploded nationwide and become a flashpoint in local and state politics. Nearly 300 bans and moratoriums on data centers have cropped up, blocking or delaying projects worth almost $130 billion this year. The AIPI poll found that 47 percent of voters would permit data centers if AI systems included safety requirements and security standards, while 38 percent favored outright bans.
The broader AI safety question has crossed into electoral politics. Democrats and Republicans in the race to replace retiring New York Congressman Jerry Nadler spent more than $40 million, with most funds flowing to groups aligned on AI regulation issues. Nearly three-quarters of AIPI respondents believe AI will become an increasingly important political issue.
A separate Pew Research survey released this month showed roughly two-thirds of Americans believe AI is advancing too fast. That poll also registered a significant partisan flip. In 2024, 70 percent of Republicans said they lacked confidence in government's ability to regulate AI, compared with 54 percent of Democrats. This month, 74 percent of Democrats expressed doubt in government's regulatory capacity, while only 61 percent of Republicans did.
Over 80 percent of respondents across both parties said AI companies should not build systems smarter than humans until they can prove they can control them.
The Trump administration's push for less regulation has collided with these polling numbers. White House officials have argued that strict rules risk slowing American AI development relative to China. Wildeford countered that claim directly: "I think people in the White House who have been trying to push a no-rules-whatsoever perspective are out of step with the American people."
Author Sarah Mitchell: "The data is clear: Americans don't trust AI companies to police themselves, and they're tired of waiting for Congress to act while regulators fumble around with voluntary frameworks."
Comments