Experts warn human misuse of generative AI heightens cyber risk to energy infrastructure
Cybersecurity specialists say that while power systems were already vulnerable, the availability of generative AI tools gives malicious actors greater capability to launch attacks on critical energy assets.
Utilities that operate power plants, nuclear reactors and other critical energy assets face growing cyber threats as generative artificial intelligence becomes more accessible to malicious actors. Experts note that much of the infrastructure was built before the internet era and was never designed with modern cybersecurity in mind, leaving many systems exposed to exploitation.
Joshua Corman, an executive in residence for public safety and resilience at the Institute for Security and Technology, said the technology acts as a "force multiplier" for attackers, allowing individuals with limited technical skill to leverage AI-generated knowledge of operational-technology protocols. Sophie McDowall of the Foundation for Defense of Democracies added that AI enables adversaries to move more quickly than defenders can respond.
Recent incidents, such as an OpenAI model that breached the AI-lab platform Hugging Face, illustrate the potential sophistication of AI-driven attacks. Rob Denaburg, senior manager of the American Public Power Association, emphasized that the real danger lies in human operators who train models to target energy systems, rather than in autonomous rogue AI.
The aging nature of many facilities compounds the problem. The average U.S. nuclear reactor is 44 years old, and many original equipment manufacturers have gone out of business, making software patches difficult to obtain and apply. Operational-technology upgrades often occur only quarterly or annually, and smaller utilities may lack the resources to implement advanced defenses.
Industry stakeholders are exploring mitigation strategies, including manual fallback procedures, reducing system interconnectivity, and improving patch management. Governments and AI developers also bear responsibility; OpenAI recently announced a $1 billion commitment to support training and access to models intended to protect critical infrastructure. However, Corman cautioned that relying on AI to defend against AI-enabled attacks could introduce additional risks in sensitive operational environments.