Written by Julien Ricciarelli-Bonnal
13 October 2026
The Essentials
Nearly two-thirds of American adults now believe artificial intelligence is developing too quickly in the United States. An AP-NORC poll published on October 8 found that 64% share that view, compared with 27% who consider the pace about right and only 8% who think AI is progressing too slowly. The finding matters because it exposes a growing gap between the logic driving the technology industry, where speed, investment and competitive advantage dominate, and the way much of the public appears to evaluate AI, with greater emphasis on human control, employment and real-world consequences. The challenge may no longer be simply to make the technology advance. Trust now has to keep up with it.
Speed has become almost synonymous with success in artificial intelligence. Each new generation of models is expected to outperform the last, every major capability becomes part of a race to launch before competitors, and companies are repeatedly encouraged to adopt AI quickly before someone else gains an advantage. The language surrounding the industry reflects this permanent acceleration: scaling, deployment, automation, autonomous agents and rapid adoption.
A large part of the American public appears to be looking at the same movement very differently. The AP-NORC survey published on October 8 found that 64% of U.S. adults believe AI technology is developing too quickly. That includes 76% of Democrats and 54% of Republicans. Only 8% of adults believe development is moving too slowly.
These figures do not mean that 64% of Americans want artificial intelligence to stop. They reveal something more interesting: much of the public does not appear to consider speed a benefit in itself. While the industry measures progress through capabilities, market position and technological leadership, people also measure it through what happens to their jobs, their personal autonomy and their ability to remain in control of decisions increasingly influenced by machines.
People are not necessarily asking for less AI, but for more control
Perhaps the most revealing finding is not the number concerned about speed. Roughly eight in ten American adults say it is extremely or very important for the government to ensure that AI remains under human control. A similarly large share says protecting U.S. workers should be a priority.
That changes how public concern should be interpreted. It would be easy to dismiss anxiety around artificial intelligence as another version of the uncertainty that accompanies every technological shift. History contains countless innovations that initially generated fear before becoming ordinary parts of life.
But that explanation becomes insufficient when people simultaneously express a much more specific expectation: human beings should retain meaningful control, and workers should not simply be treated as collateral damage in the transition.
This is closely connected to a question explored when Sam Altman said OpenAI was prepared to slow some AI development if safety concerns required it. The interesting point was not that slowing down is always desirable. It was the recognition that speed itself can become something that needs to be governed rather than treated as an unquestioned objective.
A society can welcome the benefits of artificial intelligence while still asking certain applications to progress no faster than the systems designed to supervise them. Those positions are not contradictory.
The industry and the public are using different measures of progress
For a technology company, the benefits of acceleration are relatively easy to identify. A stronger model can attract customers. Additional automation can improve productivity. Releasing a new capability ahead of a competitor can create commercial advantage. At national level, artificial intelligence is increasingly discussed through investment, economic productivity and technological competition.
For individuals, the balance looks different. The benefit may be faster research, better software, a more efficient service or several hours saved at work. The perceived downside can involve something much more fundamental: the future of a profession, the use of personal data, the reliability of an automated decision or the question of whether a human being can still intervene when the system gets something wrong.
Both sides can therefore observe exactly the same technical breakthrough and draw different conclusions from it. When an AI laboratory demonstrates that a model can now perform a task previously carried out by specialists, the industry sees progress. A person employed to perform that task may first ask what the breakthrough means for their future.
This gap cannot simply be solved with better communications. Public education about artificial intelligence matters, but rational economic concerns do not disappear because a company explains its technology more clearly. If AI genuinely changes work, responsibility and decision-making, trust has to be designed into those changes rather than added afterwards as a communications exercise.
Employment is where technological abstraction becomes personal
Predictions about AI and employment remain highly uncertain. Some occupations will change dramatically, some tasks will be automated, new roles will emerge and many of today’s dramatic forecasts will probably turn out to be wrong. Technological capability has never translated neatly into precise labour-market predictions.
That uncertainty is not necessarily reassuring to the person watching automation arrive in their profession. It can have the opposite effect. Not knowing exactly what will happen makes it harder to know which skills will retain value, which tasks may disappear and how quickly an existing role could change.
This helps explain why worker protection ranks so highly in the survey. Artificial intelligence is no longer simply a piece of software individuals choose to use. It is increasingly becoming infrastructure that can reshape a job even when the worker had no role in selecting or deploying it.
Companies should take that difference seriously. An organisation can achieve genuine productivity gains and still create internal resistance if employees experience AI as something being done to them rather than something introduced with clear boundaries, expectations and responsibilities.
The practical question is not simply how many hours can be automated. It is also what happens to the people whose work is reorganised around that automation.
Human control becomes more important as AI gains the ability to act
The issue becomes even clearer as artificial intelligence moves from generating outputs to performing actions. An AI system that drafts an email leaves a simple boundary in place: a person can decide whether to send it. An agent able to communicate externally, modify records, trigger transactions or move across multiple applications changes the nature of that boundary.
The problem is not necessarily that such systems are uncontrollable. It is that organisations must become much more precise about what the system is allowed to do independently, what requires approval and who remains accountable when something goes wrong.
That is the purpose of AI Business Governance built around use cases, responsibilities and the appropriate level of human involvement. As the tools become more capable, organisational maturity cannot be measured solely by how many activities a company automates. It also has to be measured by whether the company knows which activities should not be fully automated.
This is closely aligned with what the AP-NORC findings appear to capture. Demand for human control is not necessarily opposition to AI. It may increasingly become one of the conditions under which people are prepared to accept it.
Moving too fast can eventually slow adoption
This creates an important paradox for the technology industry. The more aggressively companies push adoption without addressing the concerns it creates, the greater the risk of generating the resistance that eventually slows adoption down.
A technology can be technically ready while the organisation around it remains unprepared. It can perform a task effectively while responsibilities are still unclear. It can generate measurable economic benefits while damaging the trust of the people expected to work with it. A rollout that looks efficient on a product roadmap may become much slower when employees, customers or regulators no longer consider the surrounding safeguards credible.
The same principle applies far beyond national AI policy. Every company adopting these tools faces a smaller version of the problem. Deploying a new capability simply because it has become available is not strategy. The business still needs to identify the problem being solved, the expected gain, the data involved, the necessary level of verification and how affected employees will participate in the change.
Technology companies naturally operate from the supply side: here is what the model can now do.
Sustainable adoption operates from a different question: here is why people are prepared to let it do it.
AI will need a form of social licence as well as technical capability
Companies understandably talk about innovation, performance and competitiveness. But technologies capable of reorganising large parts of economic life eventually encounter another variable: social legitimacy.
That does not require universal enthusiasm. No significant economic transformation receives unanimous approval. It does require enough people to understand the benefits, believe that risks are being taken seriously and feel that decisions have not moved entirely beyond human influence.
The AP-NORC findings are therefore a useful warning sign. An industry deeply concerned about the possibility of moving too slowly is facing a public in which a clear majority is expressing precisely the opposite concern.
That does not mean AI development should stop. It means acceleration can no longer be treated as sufficient justification for every deployment.
The next stage of artificial intelligence may therefore be decided by more than model performance, infrastructure investment or the number of tasks that can technically be automated. It may also depend on whether companies and institutions can answer a much simpler question: who remains in control when technological progress moves faster than public confidence?
Written by Julien Ricciarelli-Bonnal
13 October 2026
