Independent Expertise Gets a Place in the AGI Debate as 28% Choose Scientists to Judge Release Safety

A small survey points to a clear public preference: companies developing advanced AI should not have the only say over when those systems are safe to release. About 28% of respondents selected independent scientists as the best final judge, placing them alongside government regulators and the companies themselves.

A survey of 100 Americans shows how quickly the AGI debate has moved beyond questions about which company might reach the next technical milestone. Respondents were asked who should decide whether an artificial general intelligence system is safe to release, and the answers were spread across several options. Government regulators, the developing company, and independent scientific experts each drew roughly similar support, with scientists receiving about 28%.

That split matters because it shows where public confidence is headed. People appear willing to let private companies build advanced systems, but they want outside expertise involved before those systems are deployed. Technical achievement alone carries limited weight when the same company building a powerful system also controls the safety verdict.

Independent scientific experts were one of three leading choices for judging release safety. Government regulators and the developing company received comparable levels of support, while an international regulatory organization received about 12.7%.

The result does not establish a single preferred authority. It does show that many respondents expect release decisions to involve more than corporate self-assessment.

Across the responses, about 70.5% assigned the greatest responsibility to someone outside the company that developed the system. Only 29.5% gave the developer the final say. That gap provides the clearest signal in the question: the public wants an external check before a highly capable AI system reaches broad use.

Independent scientists could supply several forms of that check:

Such work would require more than a panel of respected names. Scientists would need access to relevant systems, reliable testing environments, protection from retaliation, and enough time to investigate claims. A review process that receives only curated demonstrations would offer limited protection.

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 found that 49.8% of respondents had a great deal or a fair amount of confidence that leading AI companies would put public safety ahead of the race to develop AGI. Another 50.2% had not too much confidence or none at all.

The intensity of the responses leaned skeptical. Only 15.6% expressed a great deal of confidence, while 20.2% expressed no confidence whatsoever.

That division helps explain why scientists receive serious support as release judges. A developer may understand its own system better than any outside group, but it also faces commercial pressure, investor expectations, competitive threats, and internal incentives tied to launch schedules. An independent review can add a different set of priorities to the decision.

The survey’s findings line up with a wider pattern in public opinion research. Pew Research Center reported in 2025 that 59% of Americans had not too much confidence or no confidence in US companies to develop and use AI responsibly. The concern reaches beyond AGI. It includes questions about misinformation, employment, privacy, human judgment, and the ability of ordinary people to identify AI-generated material.

For AI companies, that broader context creates a difficult environment. Every new capability arrives alongside an existing trust deficit. A company asking the public to accept a more powerful system has to address its past decisions as well as the technical details of the new model.

The survey presented respondents with a scenario in which one company appeared close to developing AGI. Only 12.4% wanted that company to continue as quickly as possible.

The other responses favored some form of restraint:

These figures point to a practical distinction. A majority of respondents still allowed room for development, but they wanted conditions attached to it. About 48.6% supported continuing development either quickly or with oversight. About 51.4% preferred slowing or stopping it.

The strongest shared preference was for a process that prevents speed from deciding the outcome. Companies can continue competing, but release decisions should depend on evidence that systems are safe enough to deploy and controllable enough to manage after deployment.

That standard presents a direct challenge to the idea that the first company to reach AGI should automatically gain the right to release it. Being first may demonstrate technical ability. It does not demonstrate that the system has passed an independent safety review, that the company can respond to failures, or that the public has meaningful protection if something goes wrong.

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 survey found that only 7.9% of respondents chose capability alone when asked to prioritize between the most capable AI system and the safest, most controllable one.

Another 44.6% considered both equally important, and 42% selected safety alone. Taken together, 86.6% placed safety on par with or above capability.

That result changes how the AGI race should be described. Many people see the technical contest, but they judge success through a wider test:

Can the system be controlled, evaluated, corrected, and held accountable after it becomes powerful?

A model that performs impressive tasks but cannot be reliably monitored presents a poor public bargain. The same applies to a company that publishes capability benchmarks while withholding details about dangerous failures, testing gaps, or limits on human oversight.

Independent scientists have a role in translating technical claims into evidence that outsiders can examine. A release review could ask whether the system’s evaluation methods match the risks created by its abilities. It could also examine how the company handles uncertainty. A credible review should identify what remains unknown instead of treating the absence of detected problems as proof of safety.

Respondents generally viewed public disclosure of serious safety failures as a trust-building step. About 47.9% said they would trust a company more if it disclosed a serious failure involving an advanced model. Another 31.1% said disclosure would make no difference, while 21% said it would reduce their trust.

The numbers give companies a useful signal. Openness about failures is more likely to help than hurt, especially among people who already have some confidence in the company. Among higher-confidence respondents, 63.3% said disclosure would increase trust. Among lower-confidence respondents, only 32.6% said the same.

That difference shows why transparency cannot carry the entire burden. Skeptical respondents may see disclosure as evidence that a problem occurred without seeing evidence that the company can prevent a repeat. A public incident report needs to connect to specific action:

Independent scientists can help answer the last question. They can review remediation plans, reproduce relevant tests, and identify whether a company has closed a real gap or simply changed the way it describes the gap.

A disclosure without independent follow-up risks becoming a communications exercise. A disclosure tied to outside review gives the public a way to judge whether the company learned from the event.