Before recent high-profile incidents involving rogue AI agents raised fears about artificial intelligence threatening humanity, the nation's energy systems were already dangerously exposed to cyberattack — and that vulnerability is accelerating, cybersecurity experts told The Verge in recent interviews. Joshua Corman, executive in residence for public safety and resilience at the Institute for Security and Technology, warned that generative AI is acting as a "force multiplier" for attackers. While some AI developers estimate a 10 percent chance artificial intelligence could one day eliminate all humans, security professionals remain more concerned about AI empowering human adversaries to strike critical infrastructure like power grids right now.
The core problem stems from infrastructure that was never built to withstand modern cyber threats. Nuclear reactors in the United States average about 44 years old, and much of the nation's critical energy equipment — responsible for keeping lights on, refrigerators running, and hospital devices functioning — predates internet connectivity and contemporary security considerations. Many companies that originally manufactured equipment still operating in the power sector have shuttered, leaving behind devices with no one to create security patches. Even when updates exist, operational technology systems controlling physical machinery for critical infrastructure might only be engineered to accept patches once per quarter or year, unlike standard IT software. Smaller utilities frequently lack the budget, personnel, and expertise to deploy the most current defensive tools.
"It's literally any sociopath that wants to [attack] is now more powerful than they used to be," Corman told The Verge. He explained that while a malicious human may lack knowledge of operational technology protocols, networks, and strategies, large language models have absorbed technical manuals and understand exactly what actions to take. Rob Denaburg, cybersecurity program senior manager at the American Public Power Association representing community-owned utilities across 2,000 municipalities, noted that recent AI agent breaches like the OpenAI model attack on Hugging Face demonstrated capabilities that were "really eye-opening and in a sense terrifying in terms of how effective they were." However, he pointed out that even rogue agents stayed focused on fulfilling their training objectives rather than deliberately targeting infrastructure. Sophie McDowall, a research associate at the Foundation for Defense of Democracies' Center on Cyber and Technology Innovation, emphasized that AI's true impact is enabling adversaries to operate more rapidly while defenders of infrastructure struggle to keep pace.
The shift represents a fundamental change in threat assessment, according to the experts. Nation-states were historically viewed as the primary danger to critical systems because of their discipline and ability to execute complex attacks. Now generative AI has lowered the skill barrier, allowing less capable attackers to mount effective assaults that previously required advanced expertise. Utilities must strengthen defenses regardless of the attacker's identity, experts say, since stopping an adversary at any single point can thwart an entire assault chain even when AI helps automate the process. Some power companies are adopting non-cyber solutions like ensuring manual operation capabilities or reducing interconnection altogether — "in the face of the AI stuff, they're starting to realize if we can't protect it, disconnect it," Corman said. McDowall argued that governments and AI developers bear responsibility too, noting that while OpenAI CEO Sam Altman's recent meeting with utilities was positive, the companies are "offering support for a problem that they are partially causing" by failing to adequately control their technology. Research into using AI to enhance energy cybersecurity remains scarce beyond basic vulnerability testing.
OpenAI announced on September 3 it would pledge $1 billion to subsidize training and access to new models designed to defend critical infrastructure, warning that "AI-enabled cyber attacks will become far more widespread and sophisticated" in coming months. But Corman cautioned against deploying friendly AI agents to battle malicious ones in sensitive operational technology environments, comparing the scenario to "an AI bull fighting another AI bull in an OT china shop" because both could prove difficult to control. McDowall called for regulatory guardrails similar to those governing nuclear technologies and hazardous materials, arguing that while research shouldn't be stifled, "there's no reason that we can't drive research forward while also doing it responsibly." The stakes are clear: a technology capable of threatening critical infrastructure and human life demands restrictions, even as development continues. The urgency compounds as utilities race to harden defenses against adversaries who grow more capable by the day, while the companies building the tools that amplify the threat have yet to fully reckon with their role in creating the problem they now promise to help solve.