AI is getting better at hacking power and water systems, new research shows - Axios

Researchers at the cybersecurity company Frenos tested 10 AI models on their ability to plan attacks against industrial systems, including software that monitors water treatment processes, Axios reported on October 21, 2026. The test assessed the models' capacity to construct attack plans rather than to carry out attacks. Axios reported the results under a headline stating that AI is getting better at hacking power and water systems.

Three of the 10 models tested successfully constructed attack plans in roughly 90% of attempts. Three models that the researchers considered at the frontier succeeded in 17.9% to 31.9% of attempts. The excerpt does not say whether any model falls into both groups, nor how many attempts each model was given. It also does not name the models tested, the companies that developed them, or state how the resulting plans were assessed.

The excerpt gives no figures for the remaining four models in the test. It does not include comment from Frenos beyond the results, from the developers of the tested models, or from operators of water or power infrastructure. Axios describes the work as new research; the excerpt does not state whether it has been published in full or reviewed by other researchers.