Leadership1 publisher3 min readPublished
Most Americans say they would stop building a dangerous AI even with rivals racing
POLITICO's new poll puts 63 percent of US adults on the view that AI could someday destroy mankind. In the same survey, 62 percent said they would abandon a project like that and expect rival labs to do the same.
The Board Room · Leadership desk
What happened
- A POLITICO survey of 2,064 US adults, fielded Sept. 13 to 15 with a margin of error of about 2.2 points, found 63 percent believe AI could someday destroy mankind.
- Dario Amodei recommended on Saturday that all AI companies adopt measures to pace development and leave room for safeguards, and Musk and Altman joined the call for a slowdown last week.
- Trump pushed back by citing competition with China and called the concern a conspiracy, and he is scheduled to meet Xi Jinping in Washington next week.
Compiled by The Board RoomSomething wrong?How this is made
Why it matters
- contradiction The labs and some lawmakers are converging on pacing while the White House treats the same concern as a conspiracy that benefits China, so a company drafting AI governance this quarter is choosing which authority it aligns with.
- constraint Twenty-six of the 63 points sit at the mildest risk level. A policy sold internally as a response to overwhelming public alarm overstates the survey, and anyone who reads the crosstabs will see it.
- exposure With half the public crediting insider safety warnings as sincere, an employee who objects publicly to a deployment arrives with a default presumption of good faith that the employer has to answer on the merits.
- precedent Amodei's recommendations are the first published industry description of what pacing means. Altman's endorsement turns them into the reference other labs and their buyers get compared against.
Twenty-six points of the 63 come from people who picked the mildest option on offer, "moderate risk", with 20 points at "significant risk" and 17 at "almost certain" [3]. Those three add to the headline figure exactly [1]. So 37 percent of adults put the risk at significant or higher [2], and the mildest tier accounts for about 41 percent of everyone counted in the 63 [3]. At a margin of error of roughly 2.2 points [1], the top-line sits between about 60.8 and 65.2 percent [4].
The figure that speaks to Dario Amodei's proposal is the hypothetical one. POLITICO asked what respondents would do if they were developing an AI product that "might destroy humanity" while knowing other countries were likely working on the same technology. Sixty-two percent said they would stop and expect other labs to do the same; 38 percent said they would keep going in the hope of heading off a more dangerous rival [4].
That second group's reasoning is the administration's. President Trump pushed back on the safety concerns by pointing to competition with China [11]. "We already have tremendous CRIMINAL and REGULATORY power over these companies!" he wrote in a Monday Truth Social post, adding: "There is a SICK conspiracy going on against AI and Data Centers, and the only one that is happy about it is China." [12] He is scheduled to meet Chinese leader Xi Jinping in Washington next week [11].
The labs are on the other side of that line. Amodei recommended on Saturday that all AI companies adopt measures to "pace" development and leave room for safeguards to be installed [6], and Elon Musk and Sam Altman joined his call for a slowdown last week [7]. "I agree with Dario that we need to pace the frontier," Altman wrote. "This has been a primary topic of discussions we've had at OpenAI in recent weeks." [8] Sen. Chris Van Hollen (D-Md.) pointed to the warning from Jacob Coxon, a former researcher at both OpenAI and Anthropic who accused the companies of "gambling with our lives" [9], as evidence it was "time to pump the brakes & take action NOW" [10].
For anyone drafting an internal AI policy, the motive question matters more than the extinction one. Asked why the lab employees went public, 51 percent said they "genuinely believe" what they are saying, 26 percent called it a "marketing move", and 22 percent did not know [5]. That is close to two sincere-belief votes for every promotional one [5]. An objecting employee starts with that presumption in place, before the employer says a word.
Intensity is soft. The mildest tier carries 26 of the 63 points [3][1], which leaves a thin base under any governance position built on the top-line. The motive split is sturdier, because it is a judgement about credibility, and it runs about two to one [5][5]. The poll did not ask respondents about AI at work, or about Amodei's measures by name.
Pacing the frontier is a decade-scale commitment for the labs. The narrower question this quarter is whether an internal policy names an external standard at all, when the only published industry description of pacing is Amodei's Saturday recommendation [6].
What to watch
- Whether Amodei's pacing measures are published as a specific, auditable list that rival labs actually sign.
- Whether next week's Trump-Xi meeting in Washington hardens the White House competition framing against any slowdown.
- A second poll re-asking the motive question, which would show whether the 51-to-26 split on insider sincerity holds.