AGI Research Puts Trust-Building in Focus as 11% Pick OpenAI as Their Most Trusted Developer

A small survey of Americans points to a large problem for the AGI industry: technical ambition is moving faster than public confidence. Respondents remain open to advanced AI, but they want safety decisions shaped by independent oversight, shared rules, and visible accountability.

A survey of 100 US adults found that only 11.4% trusted OpenAI most to develop artificial general intelligence responsibly. The same survey found that 31.7% expected OpenAI to develop AGI first. That 20.3 percentage point gap captures the central challenge facing the company and the wider industry: people may recognize technical leadership while remaining unconvinced that the leaders can manage the risks.

The findings support a mainstream public position that deserves serious attention. AGI research can continue, but a winner-take-all race needs firm boundaries. Companies will have difficulty earning durable support if they ask the public to accept powerful systems based on internal assurances alone.

OpenAI’s 11.4% trust score is significant because the company is widely associated with rapid progress in advanced AI. Respondents gave it the highest share when asked which company was most likely to reach AGI first, yet its responsible-development score was much lower.

Among respondents who predicted that OpenAI would reach AGI first, only 36% also trusted it most to develop the technology responsibly. The rest placed their trust elsewhere. About 33.1% selected Google, while other responses went to Meta, Anthropic, or another company.

Google produced a closer match between perceived capability and responsible development. Half of respondents expected Google to reach AGI first, and 54.6% trusted Google most to develop it responsibly. Google was the only company to receive majority support on the trust question.

The survey does not establish why respondents made these choices. Corporate communication, leadership controversies, commercial partnerships, safety commitments, and general brand familiarity may all influence public opinion. The numbers still matter because public trust affects adoption, regulation, and the political room available for continued research.

A company that reaches a technical milestone without public confidence may face intense scrutiny at the moment it needs cooperation. Governments may impose stricter rules. Institutions may delay deployment. Users may question whether the company has earned the authority to make decisions about systems that could affect work, education, security, and public information.

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.

The survey included 100 respondents between the ages of 18 and 63 from across the United States. Each percentage point therefore represents roughly one person in the sample. The results provide a useful snapshot of attitudes, while their size limits how confidently they can describe the entire US population.

The available material does not identify the sample as a probability sample. That detail matters. Guidance from the American Association for Public Opinion Research says that margins of error should not be reported for non-probability samples. The percentages should be read as directional findings rather than precise estimates of every American’s view.

Larger studies point in a similar direction. 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. Pew’s research involved a much larger sample and offers stronger evidence about national opinion.

The Stanford AI Index also describes a mixed public mood. More people globally said AI products and services offered more benefits than drawbacks, while a majority also said those products made them nervous. Support and concern can exist at the same time. The survey on AGI reflects that combination clearly.

The strongest finding concerned what should happen if one company appeared close to developing AGI. Only 12.4% chose the option to continue as quickly as possible.

The other respondents selected a form of constraint:

That adds up to 87.6% favoring slower progress, outside oversight, or a complete pause. The figure gives policymakers a clear signal about the public’s preferred conditions for AGI development.

The broader split is more measured. When rapid development and supervised development are combined, 48.6% favor continuing AGI development. When slowing and stopping are combined, 51.4% favor a more restrictive path.

The distinction matters. Only about one in ten respondents supported stopping development altogether. Most respondents accepted the possibility of continued research when safeguards were stronger and release decisions had independent checks.

The public’s preferred model is conditional progress: research continues, but speed alone does not determine when a powerful system is released.

That position gives companies a practical route to public support. They can keep investing in capability while showing that evaluation, monitoring, and governance have real authority over deployment.

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.

Respondents were nearly divided on whether leading AI companies would put public safety ahead of the race to develop AGI first. Those expressing a great deal or fair amount of confidence totaled 49.8%. Those expressing not too much or no confidence totaled 50.2%.

The intensity of the responses leaned skeptical. Only 15.6% expressed a great deal of confidence, while 20.2% expressed no confidence at all. This leaves companies with a narrow base of strong public trust and a larger group that sees serious reasons for doubt.

The difference between the two groups shaped their preferred policies. Among people with higher confidence in the companies, 53.1% still wanted stronger independent oversight, and only 16% supported unrestricted speed. Among people with lower confidence, 71.7% wanted development slowed or stopped.

Trust, then, does not create support for an unchecked race. It changes the preferred level of restriction. People with more confidence may accept continued development under oversight. People with less confidence are more likely to demand a slowdown or a halt.

That places a high value on institutional design. Public support will depend on whether oversight bodies can actually delay deployment, require additional testing, or demand changes when a system presents unacceptable risks.

The survey found no single winner when respondents were asked who should decide whether an advanced AI system was safe enough to release. Government regulators, independent scientific experts, and the developing company each received about 29% support. An international regulatory organization received 12.7%.

The larger pattern was clearer: 70.5% assigned the greatest responsibility to someone outside the company that built the system. Only 29.5% gave the developer final authority.

This preference supports a layered model of accountability. Government can set enforceable requirements. Independent experts can conduct technical evaluations. Companies can provide system documentation, testing data, and operational knowledge. Each group covers weaknesses in the others.

A company has the deepest access to its own model, but it also has financial and competitive reasons to move quickly. Independent evaluators can create distance from those incentives. Regulators can provide legal authority. Scientific experts can review evidence and methods without carrying the same commercial pressure.

The low support for an international organization managing AGI development does not remove the need for cross-border coordination. Advanced models operate across markets, and safety failures can travel quickly. Domestic oversight may be more familiar to respondents, while common international reporting and testing standards could still reduce regulatory gaps.

Public disclosure of serious safety failures received a generally positive response. Nearly 47.9% said they would trust an AI company more if it publicly disclosed such a failure. Another 31.1% said disclosure would make no difference, and 21% said it would reduce their trust.

The result gives companies a reason to report incidents honestly. Disclosure can show that a company is willing to accept scrutiny and learn from a failure. It can also give outside researchers and regulators information needed to identify patterns across systems.

Transparency has limits. Among respondents who already had higher confidence in AI companies, 63.3% said disclosure would increase trust. Among those with lower confidence, only 32.6% said the same. Skeptical respondents may see disclosure as useful evidence while still demanding changes to the underlying governance structure.

A credible transparency program would include more than a public statement after a crisis. It could include:

The NIST AI Risk Management Framework offers a useful reference for organizing risk management around trustworthy AI. The framework is voluntary, so companies and policymakers still need to decide which practices should become mandatory for the most capable systems.

Respondents placed safety close to or above capability when asked which goal should matter more. Only 7.9% chose capability alone. Another 44.6% said capability and safety were equally important, while 42% chose safety alone.

That means 86.6% placed safety on par with or above capability. The finding challenges the assumption that the public will tolerate major risks in exchange for more powerful systems.

For a general audience, safety can sound like a broad promise. In engineering practice, it needs measurable criteria. Companies could define how a model performs under adversarial testing, how reliably operators can monitor it, how quickly access can be withdrawn, and whether the system follows limits when placed in unfamiliar conditions.

Safety also includes control after release. A model may pass pre-release evaluations and still behave differently when integrated into a product, connected to external tools, or used by large numbers of people. Monitoring, incident response, access controls, and rollback procedures all affect whether a system remains manageable.

Independent review becomes important because companies have incentives to interpret uncertain test results in ways that support launch. External evaluators can review test design, examine evidence, and challenge conclusions before deployment decisions become irreversible.

The strongest support for a competitive structure went to companies cooperating on shared safety standards while continuing to compete. That option received 55.2% support.

Other options received much less:

When coordinated options are combined, 70.8% favored some form of cooperation or shared control. The public appears comfortable with private innovation when companies operate under common minimum expectations.

Shared standards could cover evaluation methods, incident reporting, model documentation, security testing, and thresholds for deployment. They could also reduce the advantage gained by a company that cuts safety work to release a product faster.

Cooperation will face practical problems. Companies may hesitate to share information that reveals weaknesses or helps competitors. Standards may become vague if every participant can influence the wording. Regulators and independent experts would need to check whether the rules are specific enough to measure and strong enough to matter.

The solution is a mix of voluntary coordination for routine practices and enforceable requirements for high-risk systems. A company can compete on model design and product quality while following common rules for evaluation and disclosure.

The survey gives developers several concrete priorities. Public relations campaigns may improve awareness, but trust will respond more strongly to visible changes in how safety decisions are made.

A credible program would include the following steps:

These steps will not eliminate technical risk. They can make responsibility easier to locate and give the public evidence that safety controls have operational force.