Quinnipiac Poll: 77% of Americans Want AI Development Slowed or Halted Until Safety Is Proven

A Quinnipiac University poll released this week found that 77 percent of Americans believe AI development should be slowed down or stopped altogether until its safety can be verified. The survey also recorded 86 percent support for independent safety standards governing the technology, with 74 percent of respondents expressing little or no trust in the leaders of AI companies.

The numbers land as a direct challenge to the industry's preferred approach to oversight. Major labs including OpenAI, Google DeepMind and Anthropic have largely shaped safety policy through voluntary commitments and internal review boards, arguing that rapid iteration is the only way to understand the risks. Quinnipiac's respondents appear unconvinced: three-quarters of them say the people running those labs should not be the ones certifying their own products.

Chetan Jaiswal, a researcher who studies public attitudes toward AI, framed the result as a demand for structural accountability rather than more corporate pledges. "The public wants safeguards that do not depend simply on taking company leaders at their word," he said. That distinction matters for how the survey should be read — respondents were not asked whether AI is good or bad, but who gets to decide when it is safe enough to deploy.

The findings add weight to a regulatory push already underway in several jurisdictions. The European Union's AI Act, which began phasing in obligations for general-purpose models, relies on third-party evaluation and documented risk assessments rather than self-attestation. In the United States, federal agencies have leaned on voluntary frameworks, and a handful of states have moved to write their own rules. A public that overwhelmingly favors independent verification gives lawmakers cover to tighten those requirements.

For AI companies, the poll signals a credibility problem that product launches alone will not fix. Trust in leadership sits at 26 percent, a figure comparable to the lowest readings for banks, media and Congress in similar surveys. Firms that publish third-party audit results, open their safety evaluations to outside researchers, or tie executive compensation to verified safety milestones may find they have more room to operate than those that do not.

What to watch next is whether the sentiment translates into enforcement. Polling on technology tends to be broad and shallow, and respondents were not asked how much they would pay, in higher prices or slower product cycles, for the safeguards they say they want. The test will come when a specific rule — a licensing requirement, an audit mandate, a deployment pause — reaches a vote and the industry's lobbying apparatus engages. Until then, the gap between what the public says it wants and what gets written into law remains the open question.