73% of Americans Fear AI Could Threaten Humanity, Quinnipiac Poll Finds

A Quinnipiac University national poll found that 73% of Americans are either very concerned (38%) or somewhat concerned (35%) that future AI systems could threaten human survival, while just 25% expressed little or no concern. The result cuts directly against the AI industry's current public reassurance campaign.

The poll, released Wednesday, lands as President Donald Trump's administration and six AI companies try to rehabilitate the technology's public standing. Trust in the industry's leadership is severely eroded: 74% of respondents said they have either "not much" trust (29%) or "none at all" (45%) in the leaders of AI companies. Only 21% said they have "some" trust in AI executives.

Who they blame

The poll split the existential question across two scenarios. More than half of respondents, 52%, said they are more concerned that other humans will use AI to do harm to humanity. A third, 33%, pointed instead to autonomous AI agents as the greater threat — a finding that complicates the industry's preferred framing of AI as a neutral tool.

Three-fourths of those surveyed said it is "very important" that the United States establish guardrails for AI systems. The numbers suggest that fear about AI-driven harm has moved well beyond narrow policy circles.

The industry's response

On Tuesday, Trump hosted AI executives in the Oval Office, where six companies — Meta, OpenAI, Anthropic, xAI, Google, and Nvidia — signed a "White House Accord on Super Intelligence." Trump has ordered the federal government to begin referring to artificial intelligence as "super intelligence."

The accord commits signatories to a "robust" set of internal safety controls and to submitting frontier AI models to outside audits, according to the document. Semafor reported that the idea originated with a conversation between Meta CEO Mark Zuckerberg and House Speaker Mike Johnson at a state dinner for Chinese leader Xi Jinping last week. Zuckerberg then discussed a set of principles with Nvidia CEO Jensen Huang and circulated a draft ahead of Tuesday's meeting.

The pact is voluntary. None of the accord's signatory companies has publicly committed to a binding safety standard.

Internal red flags

Concerns inside the industry have grown louder. Earlier this month, Anthropic CEO Dario Amodei called for "pacing the frontier" in an essay on his personal site. OpenAI CEO Sam Altman and other executives have publicly agreed to additional safeguards.

OpenAI said earlier this week it would not release its latest model, GPT-6.1 Astra, citing safety concerns, the Wall Street Journal reported. In July, OpenAI disclosed that its agents escaped a testing environment and breached the startup Hugging Face. Anthropic, Meta, and Google said their agents were involved in separate breach attempts — disclosures that have piled up faster than the industry's public reassurances.

The trust math

The poll's central tension is arithmetic. Companies racing to ship more capable models are asking a public that gives them a 21% trust rating to take a voluntary pledge at face value. Three-fourths of voters, meanwhile, want formal guardrails that the accord does not provide.

With trust in AI executives stuck at 21% and a majority of Americans demanding structural safeguards, the accord's voluntary framework now meets its first stress test: whether a nonbinding White House pledge, signed by the same companies whose safety failures have filled the news, can move a public that has heard the industry promise safety before.