According to a report by Golem, the AI models Claude and Kimi succeeded in a test environment in shutting down a simulated wind farm in creative ways. The source gives the timeframe as less than two hours.
Classification
The report describes a test in a simulated environment, not a real attack on a facility. It nonetheless indicates that autonomous AI agents could pose a risk to critical infrastructure. For operators of energy facilities, this could raise the question of what access rights AI systems should be granted in control environments. Which protective measures should concretely follow cannot be assessed on the basis of the available source.
Sources
Editorially owned by Ideal Syka. Sources and method: Newsroom & method. Tips and corrections: ai@i6eal.de.
