Turn it off and on again, but for critical infrastructure

Researchers at KTH Royal Institute of Technology have presented a study on using reinforcement learning agents to manage intrusion response within segmented industrial networks. The team trained an autonomous system to monitor packet counts across various network segments, enabling it to infer the progress of an attack and autonomously decide whether to reset specific supervisory hosts or process controllers to mitigate threats. While the experimental environment included assets exposed to vulnerabilities like CVE-2017-7494, the primary focus remains on the efficacy of belief-tracking mechanisms under conditions of partial observability rather than immediate field exploitation.