AGI Poll Broadens the Success Criteria as only 8% Put Maximum Capability First

A small US survey offers a clear signal about how the public may judge the race toward artificial general intelligence: capability matters, yet safety and control carry far greater weight.

A survey of 100 Americans found that only 7.9% would prioritize building the most capable AI system above every other consideration. The remaining respondents placed safety and capability on equal footing or gave safety the higher priority. That result changes the usual definition of success in the AGI race. Reaching a technical milestone first may attract attention, yet public approval will depend on whether companies can show that powerful systems are safe, controllable, and governed by people outside the companies that build them.

The survey, conducted among Americans ages 18 to 63, found broad support for continued AI development under firm conditions. Nearly nine in ten respondents rejected an unrestricted race toward AGI. Most favored slower development, independent safety oversight, or an outright pause. The findings point toward a public willing to accept ambitious research when companies accept outside scrutiny and shared rules alongside it.

The source survey included 100 people across the United States. Its sample offers a snapshot of attitudes and a useful set of questions for larger studies. It cannot establish a precise national opinion, and it should not be treated as a representative poll of every American.

That limitation matters. The American Association for Public Opinion Research advises caution when a survey uses a non-probability sample. In that situation, researchers should avoid reporting a traditional margin of error because the sample does not give every person a known chance of selection. The survey also does not provide enough information to determine how closely its demographic mix matches the US population.

The findings still deserve attention because the pattern is internally consistent. Respondents gave similar answers across several questions:

A larger, probability-based survey could test whether these results hold across the broader population. For now, the findings work best as a signal about the conditions that may shape public acceptance of AGI.

A survey of 100 Americans suggests that public approval of artificial general intelligence will depend on more than technical achievement. While respondents remain open to continued development, 86.6% placed safety at least equal to capability, and nearly nine in ten opposed an unrestricted race. The findings also reveal a gap between perceived technical leadership and institutional trust, with respondents favoring independent oversight, shared safety standards, meaningful incident disclosure, and external authority over deployment decisions. Because the sample is small and non-probability based, it offers a useful warning rather than a definitive measure of national opinion.

The survey found a sharp gap between perceptions of who might develop AGI first and which company could be trusted to develop it responsibly.

OpenAI was selected by 31.7% of respondents as the company most likely to develop AGI first. Only 11.4% trusted OpenAI most to develop it responsibly. That 20.3-point gap shows how technical reputation and institutional trust can move in different directions.

Among people who predicted that OpenAI would reach AGI first, only 36% also selected it as the company they trusted most for responsible development. The rest placed their trust in Google, Meta, Anthropic, or another option. Technical leadership gave OpenAI visibility, though it did not automatically produce confidence in its safety decisions.

Google had a closer match between perceived capability and responsibility. Half of respondents selected Google as the company most likely to develop AGI first, while 54.6% trusted Google most to develop it responsibly. Google was the only company to receive majority trust on responsible development in this survey.

The reasons behind that difference require further research. Public judgments may reflect corporate history, leadership changes, product behavior, public communication, regulatory disputes, and familiarity with each company. The survey did not ask respondents to rank those influences. Its main contribution is showing the result: a company can lead in perceived technical ability while trailing in public confidence.

That gap creates a practical risk for any AGI developer. A company with weak legitimacy may encounter greater resistance at the moment its technology becomes most consequential. Regulators may demand tougher conditions. Business customers may delay adoption. Employees and researchers may question deployment decisions. Public skepticism can become a direct operational constraint.

Respondents showed little agreement about which institution should decide when AGI is safe. Government regulators, independent scientific experts, and the company developing the system each received about 29% support. An international regulatory organization received 12.7%.

The close division among the top three choices reveals uncertainty about the best governance model. It also reveals a strong preference for outside involvement. About 70.5% assigned the greatest responsibility to someone outside the company building the system. Only 29.5% gave the developer the final say.

That response supports a layered system of accountability. No single institution has to carry every responsibility. A workable model could divide duties among several groups:

The low support for an international organization managing AGI does not eliminate the need for international cooperation. Respondents may be more comfortable with domestic accountability and expert review than with transferring final authority to a global institution. Cross-border coordination can still help with incident reporting, shared terminology, and common minimum standards.

The key point is practical: release authority needs an external check. Internal safety teams can play an important role, though their work requires independence, access to evidence, and a path to stop or delay deployment when serious risks remain.

The survey’s most important contribution is not its precise percentages but the standard it points toward: AGI developers will need to prove control, accountability, and transparency rather than simply announce capability gains. The sample is too limited to speak for all Americans, yet its consistent preference for outside scrutiny aligns with broader research showing skepticism toward companies’ ability to govern AI responsibly. Companies that treat safety as a private promise may discover that legitimacy, not performance, becomes the main constraint on deployment.

The survey also tested a specific trust-building action. Respondents were asked how they would react if an AI company publicly disclosed a serious safety failure involving an advanced model.

The ratio favors disclosure by more than two to one. Most respondents would either reward transparency or remain neutral. That gives companies a concrete reason to publish credible information about serious failures, including what happened, who identified the issue, how users were affected, and what changed afterward.

The result varies sharply by prior confidence. Among respondents who already expressed higher confidence in AI companies, 63.3% said disclosure would increase trust. Among those with lower confidence, only 32.6% said the same. Skeptical audiences may view an admission as evidence of honesty, yet they may also expect structural changes before confidence returns.

A disclosure therefore needs substance. A vague statement about an issue may create fresh suspicion. A useful incident report would address several points:

Transparency works best when it gives outsiders enough information to evaluate the response. Public relations language alone will have limited value for people who already believe companies prioritize speed over safety.

Respondents favored a cooperative structure that still allows companies to compete. A majority of 55.2% selected companies cooperating on shared safety standards while continuing to compete. Other preferences included close competition among several companies at 20.2%, government coordination at 11.2%, one company taking a clear lead at 8.9%, and international management at 4.4%.

Overall, 70.8% favored some coordinated structure. The preferred arrangement keeps private research and commercial competition in place while requiring companies to meet common safety expectations.

Shared standards could cover areas such as:

The NIST AI Risk Management Framework provides one established reference for organizing trustworthiness and risk management work. The framework is voluntary, so its existence alone does not create enforcement. A future AGI regime would need to clarify which practices become mandatory, who verifies compliance, and what happens when a company refuses or fails to meet the standard.

Cooperation also creates a coordination problem. Companies may hesitate to adopt stronger practices if competitors can move faster by ignoring them. Binding baseline rules can reduce that incentive. Each company would still compete on performance, price, efficiency, and useful applications, while the safety floor would apply across the sector.

The survey found meaningful differences by gender, age, and education. Women expressed lower confidence that companies would prioritize public safety. Only 38% of women expressed confidence on that question, compared with 61.9% of men. Women were also more likely to favor slowing or stopping development, assigning responsibility outside the developer, and using cooperative structures.

Men showed greater confidence, though unrestricted speed remained a minority position among them. Only 15.2% of male respondents supported developing AGI as quickly as possible.

Younger respondents ages 18 to 34 were more likely to view OpenAI as the probable winner and more likely to support rapid development. Even in that group, 53.2% favored slowing or stopping development. Older respondents more often placed accountability outside the developer and supported coordinated development.

Education correlated with different views of the companies. Respondents with bachelor degrees or postgraduate degrees were more likely to expect OpenAI to develop AGI first. Respondents with high school education or less showed stronger support for Google. The higher-education group also placed more weight on safety as the leading criterion and responded more positively to transparency.

These patterns may reflect differences in technology exposure, media habits, professional experience, and familiarity with corporate brands. The survey cannot separate those factors. They still matter for public communication because a single message about AGI will reach audiences with different levels of trust and different concerns.

A company that wants broad legitimacy therefore needs more than technical explanations. It needs clear evidence that can be evaluated by people with different backgrounds. Terms such as alignment, controllability, and frontier model safety require plain-language explanations tied to observable safeguards.

The 100-person survey fits a wider pattern in public opinion research, although the studies use different samples and questions.

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 same measure showed similar skepticism among AI experts, at 55%.

Another Pew study, based on 5,023 US adults, examined how Americans view AI’s effects on people and society. Its findings add context to the AGI debate: people can see potential benefits while also worrying about their ability to identify AI-generated content and about wider effects on human abilities.

The Stanford AI Index reported a global rise in the share of respondents who said AI products and services offer more benefits than drawbacks, from 55% in 2024 to 59% in 2025. At the same time, 52% said AI products made them nervous.

These results describe a public with mixed views. Concern does not automatically produce opposition to all AI research. Support for benefits does not guarantee confidence in the companies developing advanced systems. The AGI survey captures that same combination: continued development retains support, while unchecked speed receives little approval.

The survey broadens the definition of AGI leadership. The first company to reach a technical milestone may claim historical importance, yet the public will also ask how that company handled uncertainty and who had the authority to challenge its decisions.

The most important signal is the distance between capability and trust. OpenAI’s results show that the public can expect a company to lead technically while withholding confidence in its stewardship. Google’s stronger alignment between perceived capability and responsibility shows that technical credibility and institutional trust can coexist, at least in the public mind.

The path to stronger confidence runs through visible structures. Companies need to accept independent review, publish serious failures, participate in shared standards, and allow outside institutions to influence release decisions. Policymakers need to create rules that apply consistently across competing developers. Researchers and civil society groups need access to evidence that helps them assess claims.

A company that treats safety as a private promise will struggle to satisfy a public that wants proof. The survey gives that expectation a number: 86.6% placed safety at least on equal footing with maximum capability. That figure makes safety part of the success criteria for AGI, alongside technical performance.

The public has left room for progress. It has also set a condition for earning permission to proceed. AGI development will gain legitimacy when companies can show that the systems are controllable, the evidence is available, and the people making safety decisions can be held accountable.