.svg)
CISA, the FBI, and the NSA issued a joint advisory this week warning that hackers are targeting Siemens S7 programmable logic controllers (PLCs), the devices that run automated physical processes across energy, water, manufacturing, and agriculture. The advisory covers "all" S7 controllers, not a specific model or firmware version.
Attackers are using AI to generate exploit scripts, pulling from publicly available information to find and target PLCs running outdated software or otherwise poorly secured. An incident response professional who works with critical infrastructure told TechCrunch the AI use is notable, but emphasized that these devices were already highly vulnerable before AI entered the picture.
AI's role here is narrower than the headline suggests: it accelerates the search for doors that were already unlocked, at a scale no single analyst could match.
Five states, one campaign
This advisory follows a string of confirmed intrusions at water facilities in Minnesota, Michigan, Arkansas, Georgia, and New Jersey over the past several months, part of an escalating campaign CISA attributes to suspected Iran-linked hackers. CISA has warned for well over a year that these controllers should stay disconnected from the internet. Rural utilities are consistently the hardest hit, largely because they cover large service areas with small teams and even smaller budgets to maintain that separation.
The disruption risk CISA names is specific: downtime, safety incidents, and equipment damage to systems people depend on every day.
Where these compromises start
Dragos's 2025 report on operational technology and industrial control system (OT/ICS) security found that ransomware activity against industrial organizations surged more than 87% year over year, and its CEO pointed to a consistent pattern: adversaries exploiting known vulnerabilities, weak remote access configurations, and exposed OT assets to get into industrial environments. A quarter of the incidents Dragos observed ended in a full shutdown of the OT site; three-quarters caused some disruption to operations.
The PLC is the target. The remote access path, an exposed engineering workstation, a vendor VPN connection, an RDP or SSH session left open longer than it should be, is the cheap and easy way for attackers to get in. Segmenting the network and patching the controller matter. So does governing who, or what, is acting inside the sessions that reach it.
The layer underneath the perimeter
This is the layer Keystrike operates in: the post-authentication security gap between access approval and downstream detection. Keystrike governs privileged remote sessions after login, verifying that actions inside an active session come from real physical human input at the user's device, and blocking anything that does not. It helps mitigate lateral movement through interactive RDP and SSH usage, and extends visibility into non-interactive remote access paths, like PowerShell Remoting, WinRM, and WMI, that often run alongside them. When a vendor's session is covered by Keystrike, that access is governed the same way an employee's is.
Network segmentation and patch management still matter here, and so does keeping PLCs off the public internet – CISA's guidance here is clear as day. Keystrike's job sits one layer over: once a session into an OT environment is open, most tools stop asking whether the hands on the keyboard are the ones authorized to be there.
Attackers didn't need AI to find weak remote access. They needed it to find those openings faster than defenders could close them. That gap closes at the session, long before anyone reaches the controller.