NextFin News - A majority of American voters believe the Trump administration and Congress are failing to take the risks of artificial intelligence seriously, a new national poll found, even as the White House leans on a voluntary safety pact with six leading AI developers and frames restraint as a competitive liability against China. The gap between what the electorate wants — stricter government regulation, supported by 80% of Democrats and 61% of Republicans — and what Washington has delivered, a non-binding accord with no enforcement mechanism, is emerging as one of the least-discussed risks in the AI trade.
The survey of 4,506 US adults, including 3,526 registered voters, was conducted online over six days ending Monday and carries a margin of error of 2 percentage points. Its release lands less than a month before the November 3 midterm elections, with control of Congress at stake, and it puts artificial intelligence squarely in the campaign conversation alongside the economy and immigration.
What the Numbers Say
Some 57% of registered voters said the administration has not taken AI risks seriously enough or at all. That figure includes a third of self-identified Republicans, suggesting the concern is not confined to the president's political opponents. Some 54% said the same of Congress.
The skepticism runs deeper than dissatisfaction with the pace of policy. Some 62% of voters said AI could get out of control and risk the future of humankind, a share roughly unchanged since a comparable poll in August 2025. And 84% of voters see AI as a threat to American workers — a figure that towers over other perceived threats to employment. Only 60% of voters viewed illegal immigration as a threat to American workers, while 74% said the same of competition from foreign workers and 73% of large corporations. In other words, AI ranks as a bigger perceived threat to American jobs than the issue that has dominated the last two election cycles.
On the specific question of regulation, the poll found a bipartisan majority in favor of stricter government oversight of AI. Support for limits on the physical footprint of the technology is also broad: 56% of voters back limits on building AI data centers, including 69% of Democrats, 55% of independents and 43% of Republicans. Opposition to local data-center construction has already become a live issue in congressional and statewide races, with candidates from both parties calling for pauses in the current buildout.
Unease also extends to military applications. Some 57% of voters think the US government should not use AI to determine the target of a military strike, up from 49% in the August 2025 poll — an 8-point shift in a little over a year, at a time when defense planners are actively integrating autonomous systems into targeting workflows.
What Washington Has Actually Done
The administration's answer to the risk question has been a mix of promotion, rebranding and voluntary cooperation. President Trump has repeatedly voiced unhesitating support for AI and for the construction of data centers, arguing they are vital to US competitiveness. When pressed on why his government has not set limits on the technology's use and development, he often says the US tech industry will fall behind its Chinese counterparts.
"That's an important concern but it's not top of mind for most Americans, nor should it be," said Daniel Kokotajlo, a former OpenAI staffer who left the company in 2024 because he believed it was not behaving responsibly. The more pressing concern, he said, is that AI could cost people their jobs or threaten human health and safety.
In a statement on the president's handling of AI, a White House official pointed to Trump's new task force, which he said is charged with ensuring that "America continues to lead the world" on the technology.
"The Trump administration's priority is to ensure there is a booming economy so that workers have many opportunities and experience limited volatility from any short-term disruptions," the official said when asked about voters' AI-related employment fears.
The concrete policy output so far is thin. On June 2, 2026, Trump signed Executive Order 14409, "Promoting Advanced Artificial Intelligence Innovation and Security," directing agencies to harden federal infrastructure against AI-enabled risks and to work with the private sector. On June 5, National Security Presidential Memorandum-11 directed the national security enterprise to accelerate AI development and use, and to make the most advanced frontier models broadly available to national security professionals. On September 29, he issued Executive Order 14434, "Inaugurating the Era of Super Intelligence," which directs executive agencies to use the term "Super Intelligence" rather than "Artificial Intelligence" in non-statutory communications and sets a November 28 deadline for proposed legislative language defining the terms.
Last week, Trump said executives from six leading AI developers — Nvidia, SpaceX, OpenAI, Anthropic, Meta and Alphabet's Google — had agreed to a set of voluntary safety principles, including working with independent auditors to assess whether AI systems are working as their designers intended. The agreement includes no stated consequences if a company chooses not to comply. Congress has so far failed to agree on legislation that would set guardrails on the technology, and lawmakers are expected to be away from Washington until after the midterms.
The Market Is Pricing a Capex Supercycle, Not a Political Backlash
Here is where the poll matters for investors. The dominant market narrative around AI in 2026 has been an infrastructure capex supercycle: Nvidia, AMD, Arista Networks, Vertiv, Eaton and the power producers feeding the data-center boom have been priced on the assumption that buildout proceeds at pace, unimpeded. AI is the single biggest contributor to rising power demand in the United States, reversing years of gradual decline, according to strategists at Fidelity. Utilities are in a massive capital-expenditure cycle to build new generation and expand transmission and distribution lines after roughly two decades of flat demand.
But the political risk embedded in these numbers is not reflected in that valuation. When 56% of voters — and a majority of independents — back limits on data-center construction, and when 84% see AI as a threat to American workers, the assumption of frictionless deployment becomes the fragile leg of the trade. Voluntary accords signed at the White House carry no enforcement; statutory guardrails do not yet exist; and the electorate has just signaled, with unusual bipartisan intensity, that it wants them.
The transmission mechanism from ballot box to balance sheet is straightforward, and it runs through three channels. First, siting and permitting: local opposition can delay or block data-center projects, pushing out the revenue recognition timeline for equipment vendors and contractors. Second, compliance cost: mandatory audits, disclosure requirements and liability for harmful deployments raise the operating cost of frontier-model development, which favors the largest, best-capitalized labs and squeezes smaller competitors. Third, the discount rate: regulatory uncertainty is a tax on long-duration assets, and the AI infrastructure complex is the most long-duration asset class in the market — cash flows promised years out, underwritten on deployment curves that assume no friction.
Each channel raises the cost of capital and slows the deployment curve that the AI infrastructure complex is underwritten on. The risk is not that AI stops growing. It is that the growth arrives later, at higher cost, and with a wider dispersion of winners and losers than the consensus assumes. A 12-to-18-month delay in a flagship data-center campus is not a rounding error when the valuation rests on a 2027-2028 capacity ramp.
Voluntary Accords Have a Track Record — and It Is Not Reassuring
The current approach is not an innovation; it is a repetition. In July 2023, the previous administration secured voluntary commitments from seven AI companies — Amazon, Anthropic, Google, Inflection, Meta, Microsoft and OpenAI — to run red-team tests, share safety information and develop watermarking for AI-generated content before releasing systems to the public. Those commitments were hailed at the time as a meaningful step. They were also non-binding, and within two years the pace of model releases had accelerated well beyond the guardrails the companies had promised.
That history matters because it is the strongest predictor of what happens next. Voluntary regimes work when compliance is cheap and reputational. They weaken precisely when compliance becomes costly — when an audit might delay a product launch, or when disclosing an incident might spook investors ahead of a funding round. The six-company accord signed at the White House last month asks companies to do exactly those things: submit to independent audits and assess whether their systems work as intended. The incentives to perform, and the incentives to perform well, are not the same.
There is also an international dimension the administration's competitiveness argument ignores. The European Union's AI Act entered into force in 2024 and is phasing in binding requirements — risk classification, transparency obligations, and conformity assessments for high-risk and general-purpose models. China has imposed rules on generative AI that require security assessments and algorithm filings before public release. In both jurisdictions, the regulatory baseline is moving from voluntary to mandatory. If the United States remains the only major market relying on handshake agreements, the result is not American dominance; it is regulatory arbitrage, where companies optimize for the least stringent regime and the US public gets the weakest protection.
Cyclical Sentiment or Structural Regime Shift?
Is this a passing mood or a durable change in the public's relationship with the technology? The evidence points to structural.
A cyclical sentiment swing would show a sharp peak followed by decay as attention moves on, and it would be concentrated in one partisan camp. This poll shows neither. Concern that AI could get out of control is essentially unchanged from August 2025 — 62% then, 62% now — across a full year spanning an election cycle. Opposition to autonomous military targeting has moved in one direction only, from 49% to 57%. And the regulatory mandate crosses party lines: 61% of Republicans join 80% of Democrats in wanting stricter rules.
What is cyclical is the salience, not the sentiment. AI becomes a sharper campaign weapon in the weeks before November 3, and the data-center backlash flares where projects hit specific communities. Those are timing effects. The underlying skepticism — jobs, safety, loss of control, local nuisance — is anchored in material realities that will not self-correct: AI systems are being deployed into workplaces and critical infrastructure faster than norms, liability rules or auditing capacity can form. That is a regime shift in public tolerance, not a mood.
The policy gap is structural for the same reason. Voluntary industry accords have been the default precisely because Congress cannot agree on legislation; that gridlock is a feature of the current political structure, not an accident of the calendar. Unless the composition of Congress changes, or a catalyzing incident forces action, the voluntary regime is likely to persist even as public demand for binding rules hardens.
And here is the uncomfortable implication for the capex trade: a structural legitimacy deficit does not get resolved by better optics. Signing ceremonies, task forces and rebranded terminology do not lower the share of voters who believe the technology could risk the future of humankind. Only verifiable safety outcomes do — published audits, disclosed incidents, corrective actions taken. None of those exist at scale today.
The Counter-Thesis: Restraint Is the Right Call for Competition
The strongest argument against reading this poll as a mandate for regulation is the competitiveness case the administration itself makes. If the United States imposes binding guardrails while China does not, the argument runs, American developers lose the race to the most capable systems, and the economic and military upside of AI accrues elsewhere. The White House task force exists to keep America in the lead, and the voluntary accord is designed to preserve safety momentum without slowing deployment.
That argument is not frivolous, and it has supporters inside the industry. But it rests on two assumptions the poll undermines. First, it assumes the public will tolerate deployment-first policymaking; 57% of voters already say the administration is not taking risks seriously, and that includes a third of the president's own party. A government that loses public legitimacy on a technology cannot sustain a long-term deployment strategy, regardless of its industrial-policy intent. Second, it assumes voluntary cooperation substitutes for binding rules; the six-company accord has no stated consequences for non-compliance, and the 2023 precedent suggests it will not.
There is a deeper flaw in the competitiveness frame. Regulation and capability are not a simple trade-off. Clear rules — on auditing, disclosure and liability — reduce uncertainty, and reduced uncertainty lowers the cost of capital. The companies that have been asking for federal AI legislation, including some signatories of the voluntary accord, are not doing so out of altruism; they are doing so because a predictable rulebook is cheaper than a patchwork of state laws and ad hoc enforcement. The EU is discovering this the hard way as firms scramble to comply with the AI Act; the US has the chance to write the rulebook first, and is choosing not to.
The falsifying signal for the view laid out here is specific: if, after the November 3 midterms, the new Congress produces no AI legislation by the end of 2027, and data-center approvals continue at their current pace despite local opposition, then the regulatory-risk premium argued for here is overstated and the capex-supercycle thesis stands largely intact. Watch the legislative calendar and local siting decisions, not the White House signing ceremonies.
What to Watch
Short term (weeks): The midterm campaign will test whether candidates can convert AI skepticism into votes. Any competitive race where data-center opposition is a central issue is a live signal of how the sentiment translates politically. Also watch whether the six-company accord produces its first published independent audits before Election Day — a visible safety deliverable could blunt the attack line.
Medium term (months to a year): The composition of the new Congress and whether AI legislation reaches the floor in 2027. A shift in control raises the probability of statutory guardrails; a divided or status-quo Congress extends the voluntary regime. The November 28 deadline for the administration's proposed legislative definition of "Super Intelligence" is the first concrete test of whether the rebranding exercise yields substance.
Long term (years): Whether the voluntary accord produces verifiable safety outcomes — independent audits published, incidents disclosed, corrective actions taken. If it does not, public skepticism hardens further and the eventual regulatory response will be more severe, arriving after a loss of trust that is harder to repair.
Base case: a divided or narrowly controlled Congress delivers incremental, not transformative, AI legislation in 2027, and the voluntary regime muddles through while data-center opposition slows projects at the margin. Upside case for the capex trade: the competitiveness argument wins, Congress stays gridlocked, and buildout proceeds largely unimpeded. Downside case: a decisive midterm shift produces binding guardrails — mandatory audits, liability rules, siting constraints — that reprice the deployment timeline and compress multiples across the AI infrastructure complex.
The market has priced AI as an engineering and capital story. The voters are telling Washington it is also a legitimacy story — and legitimacy, once questioned, is the one input no amount of capex can buy back.
Explore more exclusive insights at nextfin.ai.
