AI, PLC security and the growing need to separate control from data
A joint advisory from five US agencies highlights how AI-assisted attacks are changing the risk profile of industrial control systems, and points to a broader question about how much direct access external systems really need to the control estate.
On 19 August, five US agencies — the NSA, CISA, FBI, Department of Energy and Environmental Protection Agency — issued a joint advisory warning of an active cyber threat targeting Siemens S7 programmable logic controllers.
What PLCs actually do
PLCs are the small, highly reliable computers at the heart of industrial control systems. They take signals from sensors, execute control logic in milliseconds and tell physical equipment what to do: open a valve, start a pump, stop a motor, adjust a pressure set-point or trigger a safety response.
What has changed is not the reliability of the controllers but the amount of data organisations now expect to pull from them.
That data increasingly has to feed historians, enterprise systems, digital twins, cloud platforms and AI applications, on top of the job the PLC was originally installed to do.
Siemens is one of the world's largest PLC suppliers, and its SIMATIC S7 family has a substantial installed base across industry and infrastructure. The advisory specifies that the threat applies across the S7 family, from older S7-200 and S7-300 models through to current S7-1200 and S7-1500 controllers.
What the agencies are warning about
According to the advisory, attackers are using internet scanning services to find S7 controllers that are exposed to the internet or poorly separated from untrusted networks.
They then use AI to accelerate the development of exploitation scripts, drawing on known vulnerabilities, weak credentials and publicly available technical information. Some of these scripts use open-source automation libraries such as snap7 and are designed to resemble legitimate monitoring software.
The technical issue centres on S7comm, the communications protocol used by Siemens engineering and monitoring tools to communicate with S7 PLCs.
If an attacker gains sufficient network and protocol access, that same legitimate mechanism can be used to read PLC memory and configuration data and, where permissions allow, write to data blocks or modify controller logic.
Ladder logic is the programme that determines how a controller responds to events in the physical process, so changing it can change how pumps, motors and valves actually behave.
The consequences they identify include disrupted industrial processes, equipment damage, safety incidents, production losses and compromised operational data.
Why AI changes the risk
The AI element matters because it changes the economics of attacking industrial systems.
AI-assisted development reduces the expertise and time needed to build working exploitation tools, allowing an attacker to gather public vulnerability information, identify exposed devices at scale and adapt scripts far more quickly than before.
Industrial environments therefore increasingly have to distinguish between legitimate software communicating with a controller and malicious software designed to use the same interfaces.
As more systems consume OT data, the number of connections, credentials and trust relationships around the control estate also grows. Each one creates another route that has to be authenticated, governed and monitored.
The architectural question underneath the warning
The agencies recommend the established foundations of OT security: accurate asset inventories, patching, removing unnecessary internet exposure, stronger credentials, proper network segmentation and DMZs, restricted engineering access and ICS-aware monitoring.
A PLC generates data that maintenance systems, digital twins, enterprise platforms and AI tools all want. That does not mean each of them needs its own direct relationship with the controller.
Reducing those relationships narrows the number of legitimate access paths that have to be defended. This matters more with every application an operator adds to an already-running brownfield estate, where the installed base is valuable precisely because it already works.
Separating control from data distribution
Altior can sit alongside an existing OT environment without replacing PLCs, rewriting their control logic or requiring a wholesale redesign of the automation architecture.
It interprets operational telemetry close to the source, normalises and structures it against the asset's digital twin, applies governance and routes authorised data to the systems that need it.
The PLC keeps doing its existing job. The data layer handles the task of making its information usable elsewhere.
Most enterprise and AI systems primarily need data rather than control access. A pump-efficiency model needs pressure, flow, operating state and history. A digital twin needs an accurate picture of the asset's current condition. A cloud platform may need telemetry from thousands of devices.
Those use cases generally do not require the same level of access as the control layer itself.
A governed operational data layer is one way to keep those functions separate, without touching functioning PLCs or forcing a control-system rearchitecture.
More data, without widening the control boundary
Altior does not replace PLC hardening, network segmentation, credential management or specialist OT security platforms.
Its value lies in helping reduce the number of legitimate paths reaching towards the control estate, while giving operational data a governed route into enterprise, cloud and AI environments.
Control can remain local, deterministic and tightly governed, while operational data moves outward through a separate layer designed for integration, cloud and AI.