AGI Poll Puts Caution in Focus as 12% Back the Fastest Possible Development

A survey of 100 Americans finds limited support for an unrestricted race toward artificial general intelligence. Most respondents would accept continued development only with stronger safeguards, outside oversight, and shared rules for competing companies.

The result is a clear warning for companies pursuing AGI. Only 12.4% of respondents said development should continue as quickly as possible when asked how companies should proceed if one appeared close to achieving AGI. The other 87.6% selected an option involving delay, independent oversight, or a full stop.

That figure matters because it puts a practical boundary around public support for advanced AI. Many people remain open to the technology and its potential benefits. They are asking companies to prove that safety decisions can withstand outside scrutiny before powerful systems reach the public.

The survey divided respondents into four choices:

The largest group favored slowing development. The second largest group supported continued work under independent oversight. Together, those two groups represent 76.8% of respondents.

The figures point toward conditional development. Respondents want researchers and companies to keep working, while requiring meaningful controls around the most capable systems. Only a small minority accepted a race in which speed sets the primary objective.

The division becomes closer when the answers are grouped by whether development should continue at all. A total of 48.6% supported continued development, either at full speed or with oversight. Another 51.4% preferred slowing or stopping it.

That split makes the public mood more precise. The survey does not show a country demanding that every AGI project end. It shows a public that sees unrestricted competition as unacceptable and wants safety conditions attached to further progress.

A survey of 100 US adults highlights a widening gap between perceived AI capability and public confidence in responsible development. OpenAI was viewed as the most likely company to develop AGI first, yet only 11.4% trusted it most to do so responsibly. Respondents generally supported continued research, but preferred slower progress, independent safety oversight, incident disclosure, shared standards, and the authority to pause risky releases. The findings are directional rather than nationally representative, but they point to a clear challenge: technical leadership will not produce durable public support without visible and enforceable accountability.

Respondents expressed almost equal confidence and skepticism when asked whether leading AI companies would put public safety ahead of the race to develop AGI first.

The intensity of those answers leans skeptical. Just 15.6% reported a great deal of confidence, while 20.2% reported none at all. The industry has a narrow opportunity to rebuild trust before more advanced systems make the stakes harder to manage.

Among respondents with higher confidence in the companies, 16% supported maximum speed. Among respondents with lower confidence, 71.7% wanted development slowed or stopped. The relationship is straightforward: confidence creates room for continued research under safeguards, while distrust produces pressure for delay or prohibition.

That puts the burden on companies. A faster model or a larger training run cannot resolve a credibility problem by itself. Technical achievement can attract attention, yet public acceptance depends on how companies make safety decisions and whether anyone outside the company can challenge them.

OpenAI shows the sharpest gap between perceived technical leadership and confidence in responsible development. In the survey, 31.7% believed OpenAI would be the first company to develop AGI. Only 11.4% trusted OpenAI most to develop it responsibly.

Among respondents who expected OpenAI to reach AGI first, only 36% also selected the company as the most trustworthy developer. The rest placed greater trust in Google, Meta, Anthropic, or another option.

Google produced a much closer match between capability and trust:

Google was the only company trusted by a majority of respondents. That alignment suggests that public confidence depends on more than a company’s ability to produce advanced systems. A company must also appear capable of controlling its own incentives and accepting scrutiny when safety concerns arise.

The survey does not identify every reason for the difference between the companies. Corporate conduct, leadership disputes, product releases, safety communications, and public controversies all shape reputation. Respondents bring those experiences to questions about AGI, even when they are asked to judge future systems.

The 15.1% who trusted another company most were especially skeptical of the current leaders. Only 4.8% of that group had meaningful confidence that leading companies would prioritize safety, and 86.1% wanted development slowed or stopped. Their answer suggests a rejection of the current industry structure more than support for one particular alternative.

The survey’s most important finding is not that the public rejects AGI; it is that people reject the idea that competitive speed should determine when powerful systems reach the world. The sample is too small to serve as a precise measure of national opinion, but its pattern is consistent with broader skepticism toward companies’ ability to govern AI responsibly. Developers should treat trust as an operational requirement, not a communications problem. Independent evaluation, transparent incident reporting, staged deployment, and real authority to delay a launch would provide stronger evidence than assurances issued by the companies themselves.

The strongest governance signal concerns who should decide whether an AGI system is safe enough to release. Three options received nearly equal support:

An international regulatory organization received 12.7%.

The larger pattern is more important than the close split among the top three options. When the answers are grouped by whether the final decision should sit inside or outside the company, 70.5% assign primary responsibility to an outside party. Only 29.5% give the developer the final say.

That is a rejection of pure self-regulation. The public may disagree over whether regulators, scientists, or a combination of institutions should have the final authority. The common point is that a company should not certify its own safety without an external check.

A workable system could involve several layers:

This structure would distribute responsibility without placing every decision in one institution. It would also give the public a visible process for handling disagreements about readiness.

Respondents were asked whether public disclosure of a serious safety failure involving an advanced model would change their trust in the company.

Disclosure therefore produced a favorable response overall. For many people, an admitted failure is easier to evaluate than a hidden one. Public reporting gives researchers, regulators, and users a chance to understand what happened and assess whether the company corrected the problem.

Trust in disclosure varied sharply by existing confidence. Among respondents who already had higher confidence in AI companies, 63.3% said disclosure would increase trust. Among the lower-confidence group, only 32.6% said the same.

That difference shows why transparency has limits. A company can publish an incident report and still face skepticism if its broader conduct suggests that the report is selective, delayed, or designed to protect a product launch. Disclosure works best when it comes with evidence of corrective action.

A credible safety disclosure should explain:

Such reporting turns transparency into an accountability mechanism. A statement that says a company takes safety seriously carries little weight without details that allow others to verify the claim.

The survey asked respondents to choose between building the most capable AI system and building the safest and most controllable system. Only 7.9% chose capability alone.

Another 44.6% considered capability and safety equally important. A further 42% chose safety alone. That means 86.6% placed safety at least on par with capability.

This preference has direct consequences for AGI development. A system that performs more tasks or reaches higher benchmark scores may look impressive, yet those gains lose public value when the system cannot be reliably monitored, restricted, or shut down.

For an average user, controllability has a practical meaning. It includes questions such as:

These questions connect technical safety work to public policy. Alignment research, evaluations, access controls, incident reporting, and model monitoring can seem abstract when described in industry terms. They become easier to understand when framed as basic conditions for responsible deployment.

When respondents were asked which competitive structure would inspire the most confidence, 55.2% chose companies competing while cooperating on shared safety standards.

Other responses included:

Overall, 70.8% supported a cooperative or coordinated structure when the options were combined. Only 29.2% preferred arrangements centered on purely competitive development.

The preferred model leaves room for private research and commercial competition. It also sets a common floor for safety. Companies could compete over model quality, cost, and useful applications while agreeing to shared testing requirements, incident reporting, and limits on dangerous capabilities.

That approach also addresses a basic coordination problem. If one company believes a competitor will move first, it may feel pressure to reduce testing or accept risks to keep pace. Shared rules can lower that pressure by making certain safeguards apply to every major developer.

The rules would need enforcement. Voluntary commitments can establish expectations, yet enforcement determines whether those expectations survive commercial pressure. Regulators and independent review bodies would need access to enough information to assess compliance.

Gender produced the largest differences in the survey. Confidence that companies would prioritize safety reached 61.9% among men and 38% among women. Women were also more likely to prefer slowing or stopping development, at 58.2% compared with 44.4% among men.

Women assigned primary accountability outside the developer at a rate of 78.4%. The corresponding figure among men was 62.4%. Support for maximum speed remained a minority position in both groups, including among men, where just 15.2% selected that option.

Younger respondents showed somewhat greater support for rapid development. Among people ages 18 to 34, 22.7% favored proceeding as quickly as possible. That group was also more likely to expect OpenAI to reach AGI first, at 39.2%.

Even among younger respondents, 53.2% wanted development slowed or stopped. Their greater familiarity with AI products may increase interest in progress while leaving substantial concern about how companies manage risk.

Education also shaped perceptions of the leading companies. Respondents with a bachelor’s or postgraduate degree were more likely to expect OpenAI to develop AGI first, at 47.9%. Respondents with high school education or less were more likely to select Google for both trust and technical leadership.

The survey cannot separate the effects of education from employment, media habits, technology use, or political outlook. The differences still matter for public communication. A single message about AGI will reach audiences with different experiences of the companies and different assumptions about who should control advanced systems.

The survey offers companies a practical agenda. Public relations alone will not close the trust gap. Companies need procedures that produce evidence of responsible conduct.

Companies should publish the criteria used to decide whether a model is ready for deployment. Those criteria should cover dangerous capabilities, misuse risks, autonomy, cybersecurity, privacy, and the ability to monitor and shut down the system.

External reviewers should be able to inspect enough evidence to test the company’s conclusions. Full public disclosure of sensitive details may create security risks, but secrecy cannot cover the entire process.

A delayed admission can make a manageable incident look like concealment. Companies should establish clear reporting thresholds and identify who has authority to publish an incident report, even when disclosure may affect a launch or stock price.

Shared standards can reduce competitive pressure to cut corners. The standards should cover evaluation methods, red-team testing, model access, post-release monitoring, and incident response.

The NIST AI Risk Management Framework offers a useful foundation for organizing risk management across the AI lifecycle. Its voluntary status limits its force, so companies and policymakers would need to decide which practices should become mandatory for the most capable systems.

An outside review group needs access to information, technical expertise, and a way to delay or restrict a release when evidence supports that decision. An advisory panel with no authority can improve communication while leaving the underlying incentive structure unchanged.

The survey points to a shift from trusting companies to govern advanced AI themselves toward requiring visible accountability.

The survey provides a useful signal, though its size limits what can be claimed. It included 100 respondents ages 18 to 63 from across the United States. The available description does not establish that the sample was randomly selected or nationally representative.

The American Association for Public Opinion Research advises against reporting a margin of error for a non-probability sample. The percentages should therefore be read as results from this group, not as a precise estimate of every American’s view.

That limitation does not erase the pattern. The support for safeguards is large enough to deserve attention, especially when it lines up with broader research. A 2025 Pew Research Center survey found that 59% of the public had not too much or no confidence in US companies to develop and use AI responsibly. The Pew study surveyed a much larger sample, making it a stronger source for national estimates.

The 2026 AI Index public opinion research also found mixed feelings about AI. Globally, 59% of respondents said AI products and services offered more benefits than drawbacks, while 52% said those products made them nervous. That combination resembles the AGI survey’s central pattern: interest in the technology exists alongside concern about its consequences and governance.

The Bakslash survey should therefore be treated as directional evidence. Its strongest use is identifying the conditions that could make continued development more acceptable: outside review, shared standards, transparent reporting, and enforceable safeguards.

The public’s position creates a demanding standard for AGI companies. Reaching a major technical milestone could establish a company as a leader in capability. It would not automatically establish that company as a legitimate authority over the system’s risks.

A company that reaches AGI first may face regulatory resistance, public backlash, or limits on deployment if users and institutions doubt its safety process. The first mover advantage becomes less valuable when the surrounding ecosystem refuses to accept the system.

The survey’s preference for shared standards offers a way through that problem. Companies can compete over research and products while accepting common rules for evaluation and release. Regulators can require disclosure and outside review without managing every research decision. Independent scientists can challenge claims without becoming the sole decision makers.

The underlying principle is simple: the more powerful the system, the more evidence a company should provide before asking the public to trust it.

The 12.4% support for maximum speed should be read alongside the 87.6% seeking constraints. It captures the public’s view of the AGI race in practical terms. Technical progress has value, but the process must show who is accountable, what happens when testing reveals a serious problem, and whether anyone outside the developer can stop an unsafe release.

Companies that answer those questions clearly can preserve room for innovation. Companies that leave them unanswered will keep turning technical competition into a public trust problem.