For a long time, the factory was considered a "closed space." Control panels remained on the factory floor, PLCs hummed quietly as part of the machinery, and HMIs and SCADA systems were meant solely for the eyes of onsite personnel. Information Technology (IT), handled by the information systems department, and Operational Technology (OT), managed on the shop floor, were separated physically, organizationally, and culturally.

This separation was not a sign of lagging behind. For a job site where operations must never stop, staying disconnected was a form of safety. Machinery untethered from networks remained untouched from the outside. It would not update without permission. It was immune to cloud outages. It would not freeze due to an OS update. Onsite personnel held the procedures deep within their muscle memory.

Thus, the fact that floppy disks still linger somewhere in a factory today is not something to be laughed at as mere nostalgia or obsolescence. It is a form of safety preserved in order to keep operations running.

Yet, into this very same site, an entirely different future is now attempting to enter. Remote maintenance, remote access, edge AI, predictive maintenance, smart factories, digital twins, next-generation networks like IOWN, cloud monitoring, and data integration across the entire supply chain. The factory is ceasing to be a closed space.

Here lies a deep distortion. The site has protected itself by staying disconnected. Management seeks to optimize by establishing connections. The issue is not the connection itself. The issue is that the design of responsibility and maintenance is not keeping pace with the speed of connection.

The boundary between factory systems and information systems is gradually shifting into ambiguity. As control equipment such as PLCs, SCADA, and HMIs connect to networks, remote maintenance and remote access become commonplace. Consequently, the attack surface widens for Industrial Control Systems (ICS)—systems that were once supposed to be contained entirely within the factory.

Yet, on many shop floors, there is no complete grasp of what is connected to what in the first place. Asset management for OT assets remains insufficient. There are legacy control panels. There are PLCs purchased online. There are devices configured by unknown hands. There are terminals missing entirely from Excel ledgers. There are connections known only to external contractors. There are procedures that exist only in the memories of veterans.

Even so, the floor keeps running. And that is precisely why the underlying issues remain hard to see.

Running does not mean safe. Not stopping does not mean managed.

Especially in small and mid-sized manufacturing, this issue grows more complex. The core of Japanese manufacturing does not reside solely in the headquarters of large enterprises. It lives in regional factories, town workshops, secondary and tertiary suppliers, component processing, prototyping, maintenance, onsite ingenuity, and machinery that has been utilized for decades. In these places, "not stopping" takes precedence over the latest AI infrastructure.

Cheap PLCs may be purchased online. In itself, this is not a bad thing. The problem is the vacuum of support that follows the purchase. Who verifies the firmware updates? Who tracks vulnerability reports? Who checks the default settings? Who designs the network segregation? Who handles recovery during a failure? Who bears responsibility in the event of a breach?

While you can buy a PLC, you cannot buy responsibility.

And here, a divide opens between headquarters and the shop floor.

Headquarters looks at the data. The shop floor listens to the sounds.

Headquarters says: Let us proceed with smart factory initiatives. Let us raise productivity with AI. Let us visualize operations with digital twins. Let us apply global standard security policies. Let us adopt Zero Trust. Let us assess the ROI as a DX investment.

The shop floor thinks: Before any of that, this machinery cannot be stopped. This PLC has been running for twenty years. This machine is understood only by that one person. Connecting to the network is terrifying. An update might break everything. A veteran's intuition can be more accurate than AI.

In foreign-affiliated companies, there is yet another layer of misalignment. The foreign headquarters issues a Global Policy. They demand a Standard Architecture. They champion One Platform. They assume Cloud-first operations. They speak of AI-driven operations. Yet, the Japanese site says: That standard does not fit this legacy equipment. Foreign sites might replace it, but in Japan, we will use it for another ten years. There is an English policy, but no Japanese operational procedure. The headquarters' AI model does not account for the exception handling of the Japanese shop floor.

What is being questioned here is not merely technology. It is a matter of whose judgment to trust. The judgment of headquarters? The factory manager? The global headquarters? The local Japanese subsidiary? The maintenance department? The information systems department? The systems integrator? The AI? The audits and certifications?

At the heart of the smart factory sits AI. It pulls data from sensors, processes at the edge, detects anomalies, predicts failures, optimizes lines, and runs simulations on digital twins. In some cases, AI judgments penetrate close to the domain of direct control.

Yet, the more central AI becomes, the more we must question not its performance, but the system of governance that determines how far we should trust it.

The digital twin is not the shop floor itself. It is a version of the floor edited through sensors, models, and inference. On the actual floor, there are elements that cannot be turned into data: the sound of a machine, the smell, the vibration, the humidity, the fatigue of the worker, the subtle variance in parts, the air of that particular day, the gut-feeling that things are "somewhat off" on the line.

Unless these elements are captured by sensors, they do not enter the digital twin. Yet, on management dashboards, the digital twin can appear more correct than the actual floor. This is where the danger lies.

Headquarters speaks of "making things visible." The shop floor feels it as "being watched."

AI can serve as a tool for rationalization. Yet simultaneously, it can serve as a tool that externalizes the judgment of those on the floor. Therefore, what is required is not merely OT security. It is a Trust OS to redesign the boundaries between IT, OT, and AI.

A Trust OS is a framework for trusting the AI. Simultaneously, it is a framework for headquarters and the shop floor to trust one another.

Is the data correct? Are the sensors calibrated? Under what field conditions was the AI model trained? How closely does the digital twin align with actual machinery and working environments? Who approves the AI's recommendations? When the AI errs, who stops it? Where is the floor's intuition channeled back? How are global standards translated for the local shop floor? What are the boundaries of responsibility for systems integrators, PLC manufacturers, cloud vendors, and maintenance companies?

One must not introduce AI alone without deciding these matters first.

We must not laugh at the floppy disk. We must not laugh at the legacy control panel. We must not solely blame the PLC bought online. We must not speak of the small and mid-sized manufacturing floor as a place lagging behind. Within it lies the wisdom of keeping things running. There is a rationality that has protected operations by staying disconnected. There is a temporal rhythm on the floor that cannot be captured by the standards of large corporate headquarters or foreign offices.

However, if we are to introduce edge AI and digital twins to that floor, we must redesign the boundaries before establishing connections. How far does IT extend? Where does OT begin? Where does the judgment of AI start? How far does human responsibility reach? Which data do we trust? Which intuitive warnings do we heed? Who holds the authority to stop operations?

What a smart factory needs is not just AI. It is a Trust OS capable of bringing the judgment of AI, the judgment of headquarters, the intuition of the shop floor, global standards, and the reality of local factories to the same table.

The future of factories does not exist solely within the digital twin. It lies in how we connect the person listening to the sound in front of a legacy control panel, the person examining data at the headquarters in Marunouchi, the person drafting standards at the global office, and the predictions generated by AI.

From floppy disks to AI. From the closed factory to the connected factory. Yet what is truly being questioned is not the evolution of technology.

What do we trust, whom do we entrust, and where do we stop?

Designing those boundaries is where the Trust OS of the smart factory era resides.

© SHIRO & Co.

First published: 2026-06-16