The AI Safety
Majority
Asked to choose among policy options for advanced artificial intelligence, U.S. voters chose mandatory safety and security standards whenever it was an option, no matter the alternative presented. The sections that follow report the structure of these preferences.
Voters choose mandatory standards over every alternative
The survey’s central items presented regulation with the counterargument attached, then asked voters to choose. They chose safety standards anyway, against every alternative offered.
Offered mandatory standards or no regulation, voters choose standards 78% to 8%.
Offered standards or a ban on more-powerful systems, they still choose standards, three to one.
Forced to pick between the two extremes alone, a temporary ban beats no rules at all. What voters decline is the unregulated option.
Voters prefer required safeguards to both no regulation and a ban
Each pairing was a forced choice between two policy positions. The standards option was specific: companies would be mandated to implement safety and security measures for their most advanced models, addressing extreme risks—explicitly, preventing AI from creating bioweapons or evading human control; the no-regulation option placed all responsibility on users; the ban would bar systems more powerful than today’s.
Against no regulation and against an outright ban, voters choose mandatory standards, 78% to 8% and 66% to 21%.
The pattern repeats on process: required review beats voluntary, 65% to 22%, and government-led standards beat company-led, 62% to 24%.
It holds for local data centers too: data centers under guardrails (covering safety, security, transparency, extreme-risk, and worker protections) beat unregulated ones 73% to 6%, and beat a flat ban 47% to 38%.
Asked whether to ban a local data center built under no AI-specific rules, most voters say ban it.
Offer the same data center with mandatory safeguards instead, and support for the ban falls nineteen points. What voters oppose is data centers without rules.
Support for oversight persists across question framings
The study asked about federal oversight of advanced AI in eight framings: five forced choices and three support-or-oppose items. Support ranges from 59% to 78%.
Support runs from 78% at its highest framing to 59% at its lowest. Oversight wins by at least five to two in every one.
Even the lowest framing, national-security testing of the most capable models, draws 59% support.
The size of the oversight majority depends on the question.
Support for keeping AI under human control
The survey tested three control requirements, framed as agree/disagree. The weakest clears 82%.
It barely moves by party: Democrats 88%, independents 86%, Republicans 83%.
84% say Congress should ensure the risk of losing control of AI is addressed, and 82% say companies should not build smarter-than-human AI until they can show it can be controlled.
Support falls only four points across the three statements, and the parties barely divide.
Partisan differences are negligible
On the flagship safety items, Democrats and Republicans land two to six points apart.
Democrats run higher on government-led standards and international coordination. Republicans run higher on the security framings.
The widest gap is export controls on AI software and hardware, 72% of Republicans to 61% of Democrats, at 11 points.
Concern outweighs excitement in every group
Asked whether they are more excited about the growth of AI or more concerned about the risks, voters choose concern, 56% to 18%. The result holds across party, age group, and gender.
Concern leads among Democrats (61% to 14%), Republicans (52% to 23%), and independents (55% to 17%) alike.
The widest split in the study is age: concern climbs from 41% among voters under 35 to 73% among those 65 and older, a 32-point spread. Even the youngest are more concerned than excited.
Preempting state AI laws is the weakest-performing proposal
Proposals to bar states from regulating AI have recurred at the federal level. Among voters, full preemption is unpopular.
Half of voters would let states act until Congress passes federal rules. Another 20% would let states set their own rules indefinitely. Together, 70% want states to keep some authority.
Full preemption peaks at 20% among Republicans, and no party or 2024-vote group gives it even one-third support.
The safety majority runs deeper than one number
Across framing, party, age, and gender, the same pattern holds: voters want advanced AI built under binding standards, kept under human control, and stoppable. The full report adds the question wording, the complete crosstabs, the methodology, and the sources behind every figure.
