Sovereign AI: Responsibility Stays In-House
Eva Mickler
7 Min. Reading time Who brings an AI model into productive operation bears responsibility for its behavior, ...
For decades, production was its own isolated world. Controls, sensors and machines ran on a network that had nothing to do with office IT. That separation was considered the best protection available: what isn’t connected can’t be attacked from afar. Digitalization, remote maintenance and the hunger for real-time data have torn down that wall. Today the factory floor often shares the same network as the data centre-and therefore the same attack surface.
Key Takeaways
Operational Technology-OT for short-is everything in manufacturing that physically moves something: programmable logic controllers, SCADA systems, sensors, robot cells, control stations. For decades this world deliberately stayed separate from the office network. No cable led out, no update came in. To tamper with a line, you had to be on site. That separation, known in the trade as an air gap, was no accident; it was a security philosophy.
The price was blindness. Nobody in the data centre saw what the machines were doing. Maintenance meant travel, data analysis meant a USB stick. Digitalization set out to erase exactly that friction-and with every step toward efficiency, the wall sprang more leaks.
Phase 1
The sneaker-net routine. Data travels via USB sticks and laptops between production and office. Networks are isolated, humans are the bridge. Stuxnet in 2010 proved even this gap offers a target for targeted attacks.
Phase 2
Remote maintenance arrives. Machine builders want to service equipment from afar, fix faults without travel. A VPN link from manufacturer to production network becomes standard. The first permanent door to the outside stands open.
Phase 3
Real-time data in the cloud. Predictive maintenance, OEE dashboards and AI analytics need a steady data stream from the plant. The controller no longer feeds just the control room but a cloud backend.
Today
One network, one attack surface. OT and IT share infrastructure, identities and pathways. A phishing email in the office can ultimately halt production. The path from keyboard to turbine is shorter than most would like.
As long as OT remained isolated, attackers needed physical access. Once separation falls, the equation changes. Now a single compromised office PC, a stolen remote-maintenance credential or a vulnerable cloud connector is enough to reach the controller. The attack surface doesn’t grow linearly-it explodes, because every new link ties the production network to the entire threat landscape of the internet.
The pattern is already documented. During the 2021 attack on Colonial Pipeline, ransomware hit IT systems, not the pipeline controls themselves. Yet operations shut down because the company couldn’t be certain how far the separation between billing IT and operational technology still held. That’s the core of convergence risk: where IT and OT intertwine, an IT incident becomes a production outage-even if no attacker ever touches a machine.
A phishing email in accounting should never be able to stop a turbine. In converged networks, it can.
Trying to secure OT (Operational Technology) with IT reflexes is doomed to fail. The two worlds operate under fundamentally opposing priorities. In IT, confidentiality comes first; a system may shut down in doubt to protect data. In OT, availability is paramount. A controller that reboots mid-batch creates scrap, and in the worst case, it costs lives.
This leads to a series of uncomfortable realities. Production systems often run for two decades or more on operating systems that haven’t received patches in years. A maintenance window for patching, as in IT, simply doesn’t exist-every minute of downtime costs real money. Many industrial protocols like Modbus or Profinet were designed for isolated networks and lack authentication or encryption. Any device on the same segment is treated as trustworthy. That assumption collapses the moment the segment starts talking to the outside world.
The blind spot
Many plants don’t even know what’s actually running on their production networks. Years of organic growth, undocumented remote-maintenance access points, and forgotten test links have created a network whose complete map no one still possesses. You can’t protect what you can’t see.
The obvious reaction-simply isolating the networks again-isn’t an option. The efficiency gains from remote maintenance and data analytics are too valuable to reverse. The task isn’t isolation; it’s controlled coupling. Three levers deliver the most impact.
The first is visibility. Before writing a single firewall rule, the production network must be fully inventoried: every device, every connection, every data flow. Passive monitoring that reads traffic without burdening the systems is the baseline. Only when you know your network can you spot anomalies-like a controller suddenly talking to an address it has never contacted before.
The second lever is segmentation following the zone model. The established Purdue Reference Model divides the factory into layers from sensor to enterprise IT and defines controlled transitions between them. Between OT and IT sits a demilitarized zone where data is exchanged without a direct path from the outside to the controller. Remote maintenance no longer runs over a permanently open VPN tunnel but through a controlled jump host that is enabled only when needed and fully logged.
The third lever is organizational-and most often underestimated. In many plants, production owns the machines and IT owns the network. No one owns the seam between them. That seam is where convergence risk lives. As long as OT security is treated as purely an IT issue or purely an equipment issue, the most dangerous zone remains ownerless.
Not another security product-an honest map. If you still can’t say with certainty which devices are on the production network and which paths they use to talk to IT or the outside world, you haven’t finished your most important homework. From that inventory, the rest follows almost automatically: secure the riskiest transitions first, convert remote-maintenance access from always-on to demand-driven, and name one person accountable for the seam between OT and IT. Convergence can’t be reversed, but it can be mastered. You just have to stop treating the shop floor like a data center. Protect it for what it is.
It describes the merging of operational technology (OT) and traditional office IT on a shared network infrastructure. Driven by remote maintenance, cloud analytics, and real-time data, machine controls today are often connected to the same network as servers and workstations. The former air gap-the physical separation of these two worlds-no longer exists in most facilities.
Because their priorities are diametrically opposed. IT systems can shut down to protect data; OT systems must keep running. Controllers often remain in service for two decades, are barely patchable, and rely on protocols without authentication. Routines like rapid updates or reboots would cripple production instead of protecting it.
Visibility before technology. First, map the entire production network: which devices are connected, what links exist, and which data flows leave the system. Without this map, neither segmentation nor anomaly detection is possible. Only then come zoning, controlled gateways, and need-based remote maintenance.
Read more on Digital Chiefs
Digital ChiefsAI deliberately eliminates middle management levelsDigital ChiefsSupply Chain Obligations Need a Data ArchitectureDigital ChiefsWhat Consultations About Transformation ConcealMore from the MBF Media Network
cloudmagazinMini-PCs displace 1U servers: Edge in the data center 2026 MyBusinessFutureTrumpf Ditzingen: Smart Factory delivers 30 percent efficiency gains SecurityTodayWhen a single call halts car productionImage source: AI-generated (July 2026)