AI-Assisted PLC Attacks on Critical Infrastructure Are No Longer Hypothetical
Five US federal agencies issued a joint warning this week confirming that attackers are actively using AI-generated code to compromise Siemens S7 Series programmable logic controllers across water, energy, manufacturing, and chemical facilities. The NSA, CISA, FBI, DOE, and EPA were unusually blunt about it: 'This is not a theoretical risk. It is an active threat.'
The attack method isn't particularly exotic, which is partly what makes it worrying. Threat actors are combining freely available industrial automation libraries, specifically snap7.dll and python-snap7, with AI coding tools to build custom scripts that masquerade as legitimate OT monitoring software. Once running, these tools interact directly with PLCs via the S7comm protocol, giving attackers read and write access to memory, configuration data, and ladder logic. That's not reconnaissance. That's control.
Targets are found the old-fashioned way: internet scanning services like Censys and ZoomEye are used to locate exposed PLCs running outdated firmware or, depressingly often, default passwords. The AI component then lowers the bar for turning a discovered device into a compromised one.
No specific group has been formally attributed in the advisory, though Iranian-affiliated actors are the leading suspects. A cyberattack in late July disrupted more than 30 community water systems across Minnesota, part of a broader pattern of intrusions targeting water and wastewater infrastructure across at least 12 US states. Cynthia Kaiser, formerly of the FBI's cyber division and now SVP at Halcyon's Ransomware Research Center, described this as 'a continuation of the same suite of activity we suspect is affiliated with Iran targeting PLCs.'
What's shifted is the toolkit. Experts had been warning that AI would eventually make its way into attacks against industrial control systems. That moment appears to have arrived. Kaiser put it plainly: threat actors are using AI to handle discrete tasks like scripting and code checks, letting them scale operations and move faster without needing deep OT expertise.
Benny Czarny, CEO of critical infrastructure security firm Opswat, made a point worth sitting with. 'AI makes it much easier for an attacker to create and modify scripts targeting PLCs, so the barrier to attacking industrial systems continues to fall. But for me the answer is not simply better AI detection.' His argument is structural: if there is a network path to a PLC from the outside world, someone will eventually use it. Data diodes, proper segmentation, and eliminating unnecessary connectivity are the actual fixes. Better anomaly detection is useful but not sufficient on its own.
The federal advisory recommends that operators immediately inventory all Siemens S7 Series PLCs, patch what needs patching, and verify that nothing is directly internet-accessible. Beyond that, monitoring for unusual S7comm behaviour is advised, including connections originating from non-engineering workstations, unexpected write operations, or sequential scanning activity on port 102.
Siemens S7 PLCs also appear in defence industrial base environments, which the agencies flagged as potential targets too.
The broader picture here isn't really about AI being some revolutionary attack multiplier. It's about operational technology environments that have been chronically under-secured, now facing attackers with progressively cheaper and faster ways to exploit that neglect. AI is accelerating the timeline, not inventing the problem.