Transparency Can Earn Trust: 48% Would Trust an AI Company More After It Disclosed a Serious Safety Failure

A survey of 100 Americans suggests that honest disclosure can improve confidence in AI companies, although transparency works best when it comes with independent oversight and enforceable safety rules.

Nearly half of respondents said they would trust an AI company more if it publicly disclosed a serious safety failure involving an advanced model. That finding offers a practical opening for companies trying to repair damaged credibility: admitting a dangerous mistake can earn respect, especially when the disclosure includes useful details about what happened and how the company plans to prevent a repeat.

The survey found that 47.9% of respondents would trust a company more after such a disclosure. Another 31.1% said disclosure would make no difference, while 21% said it would reduce their trust. In total, 79% said an admission would either improve their confidence or leave it unchanged.

That is a meaningful result in a debate often dominated by technical benchmarks, product launches, and claims about who will reach artificial general intelligence first. Public confidence depends heavily on how companies behave when systems fail. A company that publishes serious failures gives outsiders evidence that safety problems can be examined in the open.

The figure comes from a small survey, so it should be read as a signal instead of a precise national estimate. A sample of 100 people cannot represent every American, and the survey information does not establish that respondents were selected through a probability-based method. The result also does not show that disclosure alone can rebuild trust across the entire population.

Even with those limits, the direction of the response is clear. More people favored disclosure than opposed it, by a margin of more than two to one. Most respondents gave the company credit for openness or withheld judgment. That creates a strong incentive for AI companies to publish meaningful accounts of serious failures.

The details matter. A vague statement saying that a model behaved unexpectedly will carry little weight. People need enough information to understand:

A useful disclosure should help an outside reader follow the chain from failure to response. The goal is accountability that can be checked, not public relations language that simply lowers the temperature for a few news cycles.

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.

Advanced AI systems are difficult for outsiders to evaluate. Companies control much of the information about training data, testing procedures, model behavior, access controls, and incidents discovered after deployment. Independent researchers may see only selected demonstrations or limited evaluation results.

That information gap makes trust difficult. People are asked to accept claims about safety while having few ways to test those claims themselves. A public failure report changes the balance slightly. It gives regulators, researchers, journalists, customers, and affected communities material they can examine.

A disclosure can answer questions that a product announcement usually avoids:

What went wrong, how far did the problem spread, and what did the company learn before deciding what to do next?

Those questions are especially important for advanced systems that may generate harmful instructions, expose private information, manipulate users, evade safeguards, or behave unpredictably in connected tools. A company that reports such events creates a record that can be compared with future incidents.

That record can reveal whether the company is learning. Repeated incidents with the same cause suggest weak internal controls. A later report that identifies the same weakness and shows a concrete fix gives the public a basis for judging progress.

The survey showed a sharp divide between people who already had more confidence in AI companies and those who had little confidence. Among higher-confidence respondents, 63.3% said disclosure would increase their trust. Among lower-confidence respondents, only 32.6% said the same.

The difference points to a practical limit. Transparency can persuade people who see a company as capable of acting responsibly. People who already view the industry as reckless may treat an admission as evidence that the company caused a serious problem in the first place.

That reaction is reasonable. An incident report can show honesty while also confirming that a company released a system with dangerous weaknesses. Disclosure earns credibility through the quality of the response that follows. It cannot erase the underlying event.

For skeptical audiences, the strongest response would combine a public report with visible changes such as:

This approach gives people something concrete to evaluate. It also creates consequences for companies that disclose problems without addressing 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.

The disclosure finding sits within a wider pattern in the survey. Respondents were nearly evenly divided on whether leading AI companies would put public safety ahead of the race to develop AGI. About 49.8% expressed a great deal or fair amount of confidence, while 50.2% expressed little or no confidence.

The intensity leaned toward skepticism. Only 15.6% expressed a great deal of confidence, while 20.2% expressed no confidence at all. The public appears willing to consider continued development, but confidence in corporate self-restraint remains weak.

That skepticism helps explain the response to a hypothetical company approaching AGI. Just 12.4% supported developing the system as quickly as possible. The rest favored additional controls:

The preference is conditional development. Many respondents accept the value of continued research while demanding constraints on how systems are tested, released, and governed.

This is where transparency becomes part of a larger bargain. Companies can ask for room to innovate, but they need to provide evidence that their safety systems work and that outside parties can intervene when those systems fail.

The survey also found that 86.6% of respondents placed safety at least on par with capability. About 42% chose safety alone as the leading priority, 44.6% valued both equally, and just 7.9% chose capability alone.

That result challenges a common assumption within the AI industry: that users and the public will accept substantial risk in exchange for more powerful systems. The responses suggest a different standard. People judge technical progress through the conditions surrounding it.

A more capable model may attract attention, customers, and investment. A model that creates serious risks without credible controls can also invite regulatory action, deployment limits, customer departures, and public opposition. A technical lead gives a company an advantage in development. It does not automatically give the company permission to release a system.

Disclosure helps companies demonstrate that capability and safety receive serious treatment together. A company can show that its teams are tracking failures, updating evaluations, and changing deployment decisions when evidence demands it.

Respondents expressed the strongest support for a model in which companies continue competing while cooperating on shared safety standards. That option received 55.2% support. Another 20.2% favored several companies competing closely, while 11.2% preferred governments coordinating development. Only 8.9% favored one company taking a clear lead, and 4.4% supported an international organization managing AGI development.

In total, 70.8% supported some form of cooperation or coordination. The preferred arrangement preserves competition while establishing minimum rules that apply across the industry.

Disclosure fits naturally within that model. If each company reports serious incidents according to different definitions and timelines, comparisons become difficult. Shared standards could establish common categories for incidents, minimum reporting requirements, and consistent criteria for suspending or limiting a system.

A company could still compete on model architecture, product design, speed, cost, and performance. Shared safety requirements would set a floor for conduct. They would also reduce the pressure to conceal problems because competitors would face similar obligations.

The National Institute of Standards and Technology’s AI Risk Management Framework provides one reference point for organizing trustworthy AI practices. The framework is voluntary, so it does not by itself create enforceable duties. Its value comes from offering common language for identifying, measuring, and managing risk. Stronger public accountability would require translating useful practices into rules that apply to high-risk systems.

A public report can be useful without becoming a technical dump. Readers need a clear account of the event, the company’s decisions, and the remaining uncertainty.

A strong report would include these elements:

The company should explain what the system did in terms that an affected user can understand. Technical terms can be included, but they should come with explanations. The report should identify whether the behavior was an isolated test result, a repeatable model tendency, or an event that occurred in real use.

The report should state how many users, systems, or organizations may have been affected. When the company lacks reliable numbers, it should say so and explain the estimate. Readers need a clear distinction between confirmed harm, plausible exposure, and theoretical risk.

The timeline should show when the issue appeared, when the company learned about it, who escalated it, and what actions followed. Delays can matter as much as the original failure.

The company should describe the known cause and identify open questions. A report that presents every conclusion as certain will lose credibility when later evidence changes the story.

The company should specify changes to training, evaluations, access controls, monitoring, staffing, or release procedures. General promises about taking safety seriously offer little value without measurable actions.

The report should identify which outside experts, auditors, regulators, or affected groups reviewed the response. If no independent review occurred, the company should explain why.

A report should establish when the company will publish an update. Trust grows through repeated evidence, and a one-time announcement cannot show whether the fix survived contact with real-world use.

Silence may protect a company from immediate criticism, but it creates a deeper problem when failures emerge through outside reporting. A company that discloses first controls more of the factual record and can show the reasoning behind its response. A company that waits for journalists, researchers, or users to expose a problem may appear to have valued reputation above safety.

Public opinion research from Pew Research Center has also found broad skepticism about whether US companies will develop and use AI responsibly. That finding gives the disclosure question added weight. People already bring concerns about workplace disruption, misinformation, privacy, and human judgment to conversations about AI. A hidden safety failure reinforces the belief that companies will share information only when forced.

The response does not need to be theatrical. Clear facts, an accountable process, and visible correction carry more weight than dramatic apologies. Companies can acknowledge uncertainty without surrendering control of the situation. They can explain a difficult technical issue without excusing weak decisions.

The survey’s 47.9% result offers encouragement, but it also sets a demanding standard. Almost half of respondents could become more trusting after a disclosure. That opportunity disappears when disclosure arrives late, omits important details, or lacks any independent check.

AI companies seeking public acceptance of more powerful systems should build reporting into normal operations. Serious failures should trigger defined internal steps, external notification, public communication, and follow-up review. These procedures should exist before a crisis, when decisions can be made calmly and responsibilities are clear.

The public’s position is practical. Americans remain open to AI development, but they expect safety decisions to be visible and enforceable. They want companies to compete within shared boundaries, and they want independent people and institutions to have a meaningful role in judging risk.

A company that discloses a serious failure gives the public evidence that accountability can begin. A company that combines disclosure with outside review, stronger controls, and measurable correction gives that evidence a chance to become trust. Technical leadership earns lasting legitimacy only when the safety decisions behind it can withstand scrutiny.