AI-generated attack scripts are probing Siemens PLCs

Attackers no longer need years of OT experience to reach a programmable logic controller. They ask an AI model for the script.

That is the core of a joint advisory five US agencies published on August 20. In AA26-231A, CISA, NSA, FBI, the Department of Energy and the EPA warn that threat actors use AI to generate exploitation scripts against internet-exposed Siemens S7 controllers, from the retired S7-200 to the current S7-1500. The agencies call it an active threat.

What the advisory describes

The scripts are plain Python built on open-source libraries such as python-snap7. They speak the S7comm protocol and give read and write access to controller memory, configuration data and ladder logic, the program steering the physical process. Some are disguised as ordinary monitoring utilities, so they blend into normal engineering traffic.

Finding targets takes no more skill than writing the scripts. The actors locate exposed controllers through internet scanning services like Censys and ZoomEye, then log in with default or weak credentials. So far the agencies describe reconnaissance and capability building. The step they expect next is write operations meant to disrupt the process itself.

The sector list reads like a map of Norwegian industry: manufacturing, energy, water and wastewater, chemical, food production. S7 controllers run production lines and pumping stations here too, and a Censys query does not stop at a border.

Why this warning lands harder than the last one

Exposed industrial controllers are an old problem. What changed is who can exploit them. Working against S7comm used to demand niche protocol knowledge, and that scarcity kept most opportunists out. Now a model drafts a working S7 client in minutes, and OpenAI has just shipped a restricted model built specifically for exploit work. The pool of people who can go after a PLC grew from a handful of specialists to anyone with a chat window.

Our advice is to treat this as an exposure question before a malware question. Three moves, in order:

1. Learn what answers from the internet. Scan your public IP ranges and check whether anything responds on TCP port 102, the S7comm port, or any other OT protocol. You see exactly what an attacker’s Censys query sees.
2. Close what you find. Move controllers behind a firewall or VPN, and replace every default credential. The advisory names weak and default passwords as the entry point, so this single change removes the access path the scripts rely on.
3. Watch the traffic that remains. In our exposure management work we map which assets can reach the OT segment at all. Tenable OT Security detects and alerts on unauthorized S7comm reads and writes, so a rogue script shows up before it touches ladder logic.

The agencies wrote the advisory for American utilities. The scanning services it describes index Norwegian address space with the same enthusiasm.

Talk to Anders Helgesplass if you want to know what answers from your IP ranges before someone else’s model finds out.

Drafted with AI assistance, reviewed and edited by Anders Helgesplass and the FM CyberSecurity editorial team.

Sources

• CISA, NSA, FBI, DOE, EPA, “Defending Against an Active Threat to Siemens S7 Series PLCs,” advisory AA26-231A, August 20, 2026
• Help Net Security, “US agencies warn of AI-powered attacks on Siemens industrial controllers,” August 20, 2026
• The Hacker News, “AI-Generated Exploit Scripts Target Siemens S7 PLCs in U.S. Critical Infrastructure,” August 2026