Attackers are using AI-assisted scripts to target industrial controllers made by Siemens, according to a joint US government warning showing how generative tools can make longstanding operational-technology weaknesses easier to exploit without requiring a newly discovered software vulnerability.

The advisory covers Siemens S7 programmable logic controllers used across manufacturing, energy, water, chemicals, food production, agriculture, and other industrial environments. US security and infrastructure agencies describe active reconnaissance and capability development against poorly protected installations.

Attackers are combining publicly available information with open-source industrial automation libraries and AI-assisted scripting. Their tooling can mimic legitimate monitoring software and interact with controller memory, configuration data, and the ladder-logic programmes used to manage physical processes.

For industrial operators, the difference between a new vulnerability and a new way of exploiting familiar weaknesses is important. Siemens said it had not detected previously unknown vulnerabilities or an increased level of attacks against its industrial-control products, while the government guidance concentrates on internet exposure, outdated software, weak credentials, poor segmentation, and known flaws.

AI accelerates an old industrial problem

Programmable logic controllers sit much closer to physical operations than ordinary business software. They can control pumps, motors, production machinery, treatment processes, and safety-related equipment, meaning unauthorised access can affect uptime, product quality, equipment condition, and potentially human safety rather than simply exposing information.

The agencies describe the activity as “not a theoretical risk — it is an active threat”. They say AI-assisted development is reducing the expertise and time required to create usable industrial-control scripts while allowing attackers to collect vulnerability information, find exposed devices, and adapt tooling more rapidly.

The techniques described include internet scanning, weak or default credentials, open-source automation libraries, and attempts to read or alter controller data. None began with generative AI, but automation changes how cheaply and quickly they can be combined.

Industrial systems have often remained in operation for many years, while remote maintenance, systems integration, cellular connectivity, and convergence between corporate and operational networks can create access routes that were never part of the original plant design.

AI changes the economics of attacking those environments by helping less specialised actors assemble or modify tools more quickly. The expertise needed to understand a physical process, evade detection, and create a particular operational effect remains substantial, but generating scripts that interact with known industrial protocols is becoming a lower barrier.

The warning reaches European infrastructure

Although the current advisory concerns activity observed against US installations, Siemens S7 controllers are deployed globally and the manufacturer is headquartered in Germany. European utilities including Verbund and E.ON have confirmed use of the technology while stressing that relevant systems are isolated from direct internet access.

Those responses illustrate why asset exposure has become a resilience issue rather than a narrow patch-management task. Operators need to know which controllers exist, which software versions they run, which engineering workstations can reach them, how contractors gain access, and whether unexpected network routes have made a device externally reachable.

The joint advisory recommends inventories of S7 controllers, application of relevant patches, stronger access controls, monitoring for anomalous activity, and strict separation from the public internet. It also advises organisations to inspect network paths used by industrial protocols and separate operational technology from corporate networks.

That approach resembles the broader shift in European security policy. The EU’s AI cybersecurity action plan recognises that advanced AI can increase the speed of offensive activity while giving defenders additional automation capabilities.

The industrial setting makes that contest uneven. A compromised laptop can often be rebuilt comparatively quickly, whereas changing firmware, segmentation, or authentication around production equipment can require shutdown windows, vendor coordination, testing, and safety review.

Legacy machinery may remain economically productive long after the assumptions behind its original network design have become obsolete. That creates a remediation problem measured in physical operating constraints rather than ordinary software release cycles.

The emerging risk is therefore not that AI has discovered a secret route into Europe’s factories and utilities. It is that attackers can automate exploitation of weaknesses defenders have known about for years, reducing the cost of finding exposed systems while operators still have to remediate physical environments carefully, one installation at a time.