Dozens of U.S. Officials Say There’s at Least a 10% Chance Humans Lose Control of AI

Alarm Bells Inside the Pentagon

A new survey of current and former U.S. government officials has revealed a striking level of anxiety about advanced artificial intelligence: dozens of respondents believe there is at least a 10 percent chance that humanity will lose control of AI systems. The findings, released Wednesday, capture a growing unease within the very institutions tasked with managing the technology’s military and national security applications.

Nearly half of the surveyed officials serve or have served in the Department of Defense, an agency that has been locked in fierce internal debate over how to test, deploy, and govern AI without creating catastrophic risks. The results underscore how the conversation about existential AI risk has moved from the fringe of academic debate into the heart of the U.S. security establishment.

Quantifying the Unthinkable

The survey, conducted by Nextgov/FCW, asked current and former officials to estimate the likelihood that humans could lose control of advanced AI systems. While the exact sample size and methodology were not detailed in the immediate release, the fact that dozens of respondents placed the probability at 10 percent or higher is itself significant. In risk assessment terms, a 10 percent chance of a civilization-altering event is not a trivial number—it is the kind of figure that demands rigorous safety protocols and contingency planning.

The 10 percent threshold is especially notable because it separates casual concern from a more concrete, measured belief that the danger is real and plausible. This is not a fringe view; it is a viewpoint held by a sizable minority of people who have shaped defense and technology policy.

Defense Department at the Center of the Debate

The disproportionate representation of Defense Department personnel in the survey highlights the Pentagon’s central role in the U.S. government’s AI strategy. The Department of Defense is aggressively pursuing AI for everything from autonomous drones to battlefield decision-support systems. At the same time, internal watchdogs and safety advocates have warned that rushing to deploy AI without adequate guardrails could lead to unintended escalation, loss of human oversight, or even systems that behave in ways their designers cannot predict.

The survey results come amid renewed congressional scrutiny of military AI programs and a push from the White House Office of Science and Technology Policy to establish clearer safety standards. The National Institute of Standards and Technology (NIST) has been developing an AI Risk Management Framework, but translating that framework into binding rules for defense systems remains a politically and technically fraught challenge.

Existential Risk vs. Immediate Governance

It is important to distinguish between the kind of long-term, existential risk that the survey probed and the more immediate governance issues that dominate daily policy debates. The 10 percent figure reflects a belief in a future where AI systems could become so powerful or autonomous that human control is permanently lost. That is a far cry from today’s concerns about biased algorithms, data privacy, or even autonomous weapons that operate within constrained rules of engagement.

Nevertheless, experts argue that the two are connected. A governance framework that fails to address worst-case scenarios may prove inadequate as AI capabilities accelerate. The survey suggests that many officials see the current trajectory as dangerously short-sighted.

“When you have a significant number of people inside the national security apparatus saying there’s a one-in-ten chance we lose control, that’s a signal that can’t be ignored,” said one former Pentagon official who was not involved in the survey. “It means we need to invest much more seriously in alignment research and international cooperation.”

Who Was Surveyed and How?

The article from Nextgov/FCW noted that the survey targeted current and former U.S. officials with exposure to AI policy. The pool included individuals from the Defense Department, the intelligence community, and civilian agencies. The survey did not distinguish between respondents answering in their official capacity or as private citizens, which leaves some ambiguity about the institutional weight of the findings.

Still, the fact that nearly half of the respondents have a Defense Department background means the military’s internal culture is a powerful lens through which the results should be interpreted. The Pentagon has its own history of grappling with existential risk—nuclear weapons command and control—and that experience may make its personnel more attuned to low-probability, high-consequence scenarios.

Policy Implications

The survey is likely to fuel ongoing debates in Washington over how quickly the U.S. government should move to regulate frontier AI models. Some lawmakers have proposed mandatory safety testing and licensing for the most powerful AI systems. Others want to see an international treaty governing autonomous weapons. The survey’s findings could give new ammunition to those arguing that the risk of human loss of control is not hypothetical but a real concern shared by the government’s own experts.

At the same time, the results may intensify the Pentagon’s internal tug-of-war between those who advocate for rapid AI adoption to maintain a military edge and those who argue for a slower, more cautious approach. The Department of Defense has so far resisted calls for a blanket moratorium on autonomous weapons, but a growing chorus of internal voices warning about loss of control could shift that calculus.

The survey also raises questions about public transparency. If dozens of current and former officials privately believe there is a significant chance of losing control, what does that mean for the American public—and for allies who rely on U.S. leadership in setting global AI norms?

As the government continues to pour billions into AI research and development, the survey serves as a stark reminder that the people building and overseeing these systems are far from certain about where the technology is heading.