Safeguards Matter in the AGI Race: 41% Would Slow Development Until Stronger Protections Exist

A small survey of 100 Americans offers a clear warning for companies pursuing artificial general intelligence: speed alone will not earn public support. About 41% would slow development until stronger safeguards are in place, while nearly nine in ten favor some form of constraint on an unrestricted AGI race.

The result deserves attention because it captures a practical concern that reaches beyond technical debates. People may accept continued AI development, yet they want safety decisions backed by independent review, public disclosure, and rules that companies cannot rewrite for themselves. Technical leadership carries limited value when the public cannot see how a company handles serious risks.

The survey found that 40.5% of respondents preferred slowing AGI development until stronger safeguards are in place. The figure rounds to 41%, which gives this article its headline. Another 36.3% supported continued development with stronger independent safety oversight. A further 10.8% favored stopping development entirely.

Only 12.4% chose the option to continue developing AGI as quickly as possible. Taken together, 87.6% selected slowing, oversight, or a halt over an unrestricted race.

Those figures describe a public that is willing to tolerate continued research under conditions. The preferred conditions vary, but the direction is consistent: companies should have to meet clear safety expectations before moving from research to increasingly powerful deployment.

The survey’s central message can be stated plainly:

This position supports innovation with limits. It gives companies room to build and test systems while placing boundaries around the decisions that could affect the public at large.

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 respondents were almost evenly split on whether leading AI companies would put public safety ahead of the race to develop AGI. 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 concern tilted more sharply. Only 15.6% expressed a great deal of confidence, while 20.2% expressed no confidence at all. That distribution suggests that skepticism is more concentrated than strong support for corporate self-governance.

The difference matters for companies planning public rollouts. A person who has some confidence in AI development may still demand outside review. The survey found that even respondents in the higher-confidence group showed little enthusiasm for unrestricted speed. Only 16% of that group supported moving as quickly as possible, while 53.1% still wanted stronger independent oversight.

Among respondents with lower confidence, 71.7% wanted development slowed or stopped. Every respondent who selected stop development entirely came from that skeptical group.

The public therefore appears persuadable, but reassurance alone will have limited reach. Trust will depend on visible changes in how companies test systems, report failures, and make release decisions.

The survey produced a striking result for OpenAI. Respondents were most likely to identify OpenAI as the company that would develop AGI first, at 31.7%. Only 11.4% trusted OpenAI most to develop AGI responsibly.

That 20.3 percentage point gap separates perceived technical ability from perceived judgment. Among people who expected OpenAI to win the race, only 36% also trusted it most to handle development responsibly. The rest placed their trust elsewhere or selected another option.

Google received a different response. Half of respondents expected Google to develop AGI first, and 54.6% trusted Google most to develop it responsibly. Google was the only company to receive majority trust for responsible development in the survey.

The figures do not prove why respondents made those choices. Corporate reputation, public communication, leadership changes, product behavior, regulatory disputes, and familiarity may all influence perceptions. The broader lesson remains useful: a company can appear technically advanced while still struggling to establish legitimacy.

A powerful system requires more than a credible demonstration of capability. The public also wants evidence that the people controlling the system will accept delays, independent criticism, and outside intervention when the risks justify them.

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 did not agree on one institution that should decide when AGI is safe. Government regulators, the developing company, and independent scientific experts each received about 29% support. An international regulatory organization received 12.7%.

The larger pattern is easier to read. About 70.5% assigned the greatest responsibility to an actor outside the company that developed the system. Only 29.5% gave the developer final authority.

That preference creates a direct challenge to pure self-regulation. Companies have technical expertise that regulators and outside reviewers may lack, but technical expertise does not settle questions about acceptable public risk. A developer also has commercial incentives tied to launch timing, market share, and investor expectations.

A workable model could divide the responsibilities:

This approach would avoid placing every decision in the hands of one institution. It would also make it harder for a company to certify its own system without meaningful challenge.

The NIST AI Risk Management Framework offers a useful reference point because it organizes trustworthiness around design, development, use, and evaluation. Its voluntary nature also highlights the next policy question: which practices should remain recommendations, and which should become mandatory when systems reach higher risk levels?

Nearly half of respondents, 47.9%, said they would trust an AI company more if it publicly disclosed a serious safety failure involving an advanced model. Another 31.1% said disclosure would make no difference, and 21% said it would reduce trust.

The result gives companies a practical reason to report incidents. Disclosure was more likely to improve trust than damage it, especially among respondents who already had some confidence in the industry. Among the higher-confidence group, 63.3% said disclosure would increase trust. Among the lower-confidence group, 32.6% said the same.

For skeptics, a public statement may carry little weight if the company continues operating under the same incentives. An incident report becomes meaningful when it includes details about the cause, the systems affected, the harm observed, and the changes made afterward.

A useful disclosure should answer several questions:

A company does not need to publish sensitive security information that would make exploitation easier. It can still provide enough detail for outside experts, users, and regulators to judge whether the response matches the seriousness of the event.

Transparency earns credibility when it changes behavior. A disclosure without corrective action can reinforce public suspicion rather than relieve it.

When asked to choose between the most capable system and the safest, most controllable system, 42% selected safety alone. Another 44.6% said both mattered equally. Only 7.9% selected capability alone.

In total, 86.6% placed safety on par with or above capability. That preference explains why the unrestricted race receives such little support. People generally understand that more capability can make a system more useful, but they also recognize that advanced systems can create larger consequences when their behavior is unreliable or difficult to control.

The word safety covers several technical and institutional concerns. For an advanced model, it may include resistance to dangerous instructions, protection against misuse, predictable behavior under stress, secure tool access, reliable monitoring, and the ability to restrict or shut down the system when necessary.

Controllability also depends on deployment choices. A model operating in a limited research environment creates different risks from one connected to financial systems, medical decisions, public infrastructure, or autonomous software tools. Safety claims therefore need to describe the setting in which a system will operate.

Capability evaluations alone cannot answer those questions. A model may perform well on benchmarks and still create risks through unexpected combinations of tools, access, incentives, or users. Independent testing should examine the full deployment environment rather than focusing only on a model’s headline score.

The strongest support for a governance model went to companies cooperating on shared safety standards while continuing to compete. That option received 55.2%. Government coordination received 11.2%, while 8.9% favored one company taking a clear lead. Only 4.4% favored an international organization managing AGI development.

When the cooperative and coordinated options are combined, 70.8% favored some form of shared structure. Purely competitive arrangements received 29.2%.

This is a significant distinction for policymakers and executives. The public does not appear to be asking for a single government-run AGI program. It is asking companies to compete inside a common safety floor.

Shared standards could cover areas such as:

The standards would need enforcement to matter. If companies can quietly abandon them whenever a competitor moves ahead, cooperation becomes a public relations exercise. Regulators could require compliance for certain classes of systems, while companies could publish additional voluntary commitments above the legal baseline.

The survey found meaningful differences by gender. Women expressed less confidence that companies would prioritize safety, with 38% expressing confidence compared with 61.9% of men. Women were also more likely to favor slowing or stopping development, at 58.2% compared with 44.4% of men.

Women were more likely to assign accountability outside the developer as well, at 78.4% compared with 62.4% of men. Even among men, support for unrestricted speed remained limited: 15.2% favored developing AGI as quickly as possible.

Age differences were smaller. Respondents ages 18 to 34 were more likely to identify OpenAI as the likely first developer and more likely to support rapid development. A majority of younger respondents, 53.2%, still preferred slowing or stopping development.

Education also influenced company perceptions. Respondents with bachelor’s or postgraduate degrees were more likely to expect OpenAI to develop AGI first, at 47.9%. Respondents with a high school education or less were more likely to trust and select Google.

These differences may reflect media habits, professional experience, familiarity with products, and different assumptions about technological risk. They also suggest that a single public messaging strategy will reach different groups unevenly. Companies that want broad support will need evidence that holds up across audiences, rather than appeals aimed only at technology enthusiasts.

Several common interpretations can weaken the value of public-trust data and lead to poor AGI governance decisions.

The Bakslash survey included 100 Americans ages 18 to 63. Its results provide a snapshot of opinion and identify clear patterns, but the sample size is small. The available material does not establish that the respondents were selected through a probability-based method or that the sample represents the national population.

That distinction matters. AAPOR guidance notes that non-probability surveys should not report a conventional margin of error because respondents do not have known, equal chances of selection. The percentages should therefore be read as signals from this sample rather than precise national estimates.

Larger research points in a similar direction on several issues. Pew Research Center reported that 59% of the public had not too much or no confidence in US companies to develop and use AI responsibly in a 2025 survey. Stanford’s 2026 AI Index reported a rise in the share of people who saw more benefits than drawbacks from AI products, alongside a rise in the share who said AI made them nervous.

The exact percentages differ because the studies ask different questions and use different samples. The broader pattern is stable enough to guide a serious policy conversation: interest in AI can coexist with concern about who controls it and how risks are managed.

Public trust will depend on actions that people can inspect. A company that says safety comes first should be able to show how that priority affects launch decisions, budgets, staffing, contracts, and executive incentives.

Several tests would make those claims more credible:

These measures carry costs. Independent testing can delay releases, expose weaknesses, and give competitors information about a system. Documentation takes time, and public incident reporting can create legal and reputational risk.

Those costs are part of responsible development. A system with broad social consequences should not be governed by the same release habits used for a low-risk software update. The companies that build advanced systems will need to treat delay as a legitimate safety decision rather than as evidence of failure.

The survey’s 41% figure matters because it reflects a concrete preference. Respondents want development slowed until stronger safeguards exist. The 36.3% who favor independent oversight add another large group that accepts continued work only with outside checks.

A public standard for responsible AGI development could begin with four requirements:

The survey also shows why a winner-take-all approach is unlikely to build lasting legitimacy. A company may reach an important technical milestone first and still face resistance from users, workers, regulators, and public institutions if it cannot demonstrate independent and enforceable safety decisions.

AGI development can retain room for ambition and competition. The public is asking for a process that earns permission to move faster through evidence, oversight, and responsible conduct. That is a reasonable condition for systems expected to influence nearly every part of society.