As manufacturing moves from isolated automation towards connected, intelligent and AI-driven factories, the cybersecurity equation is changing. Operational technology (OT), once largely contained within the plant and separated from enterprise IT, is increasingly being connected to networks, cloud platforms and analytics systems.
For Ambarish Singh, CISO, Godrej Enterprises Group (GEG), that convergence creates significant opportunities but also a risk that manufacturers cannot afford to underestimate.
“OT systems, when they are designed to be connected, were built for a specific purpose,” Singh says. The difference today is that manufacturers increasingly want to extract data from those systems for predictive maintenance, analytics and AI.
Traditionally, OT operated with an air gap from IT. That separation limited the routes through which a cyber incident could move. But as sensors and connected systems bring plant data into IT environments and cloud platforms, the gap between the two worlds is becoming smaller.
The benefits are clear. Data from machines can be analysed to predict failures, improve uptime and optimise productivity. But Singh argues that manufacturers must secure the bridge through which that data travels.
“If you don’t do it, the impact is very high,” he says, pointing to lack of awareness and visibility into OT environments as major challenges. Manufacturers need to understand what they are integrating, why they are integrating it and what security controls are being put in place so that only the required traffic moves in or out.
The OT breach is different
The consequences of an OT compromise can extend far beyond data loss.
In an IT breach, organisations may face compromised data, systems or services, followed by recovery and restoration. An OT breach can stop a production line altogether. In certain industrial environments, the consequences can potentially extend to safety and human life.
A plant shutdown can result in operational and revenue impact, while a sufficiently serious OT incident can create physical safety risks.
That changes how security has to be designed. The objective is not simply to protect information but to protect the availability and safe operation of physical processes.
One-way traffic can be useful where the requirement is primarily to extract information from OT environments. Singh points to technologies such as data diodes as an option for securing such flows, limiting unnecessary movement between environments.
Remote access remains a weak point
Remote access is another persistent challenge. Manufacturing plants often depend on OEMs and external partners for maintenance and troubleshooting, sometimes across geographies.
The temptation, Singh says, is to provide direct internet access to an OT system to allow an external party to connect. That approach can expose the plant to unacceptable risk.
“Connecting an OT system on the internet is suicidal,” he says, warning that an exposed system can be compromised rapidly.
The challenge becomes even harder because OT environments cannot always be patched in the same way as IT systems. Industrial equipment often has long lifecycles, and replacing a controller or machine simply because its software is outdated may not be commercially or operationally feasible.
Singh describes machines running on older operating systems such as Windows XP or Windows 7, where the underlying controller and software are tightly coupled to an expensive industrial machine. “You cannot replace this easily,” he adds.
This is driving manufacturers towards compensating controls such as virtual patching, which can provide protection while organisations plan a more permanent upgrade.
Segmentation becomes the first line of defence
For legacy OT that cannot be upgraded or replaced, Singh emphasises defence in depth.
“Network segmentation” is a key starting point. The objective is to make it harder for an attacker who gains access to one part of the environment to move laterally across the plant.
That means multiple layers of access control and firewalls, as well as an industrial DMZ for manufacturing environments. Singh also cautions against simply deploying identical security infrastructure everywhere. Different layers and technologies can make it harder for a compromise of one control to undermine the entire environment.
Technology, however, is only one part of the equation. People working on the shop floor may have spent years operating machines and may not view cybersecurity as part of their role.
“Awareness is important,” Singh avers. Security controls therefore need to be accompanied by awareness, clear responsibilities and processes that fit naturally into plant operations.
AI introduces a new security layer
The arrival of AI makes the question more complex. AI systems are increasingly being positioned to support quality, predictive maintenance, optimisation and other manufacturing decisions. That means the security discussion can no longer stop at the network or machine.
Singh identifies three areas that need to be considered separately: securing the AI system itself, protecting the data on which it is trained, and controlling the actions the AI may ultimately be authorised to take.
The third is particularly important in manufacturing. “Who is taking decisions, who is actually working? That has to be fully vetted by the OT people,” he points out. Plant personnel understand the physical process and what could go wrong if an AI-generated recommendation or action crosses an operational boundary.
An AI model trained on poisoned or manipulated data could potentially generate unsafe recommendations. As AI moves closer to operational decision-making, manufacturers therefore need governance around the model, its data and its outputs, rather than treating AI security as simply another IT application-control problem.
Security cannot create friction on the shop floor
For a factory operating 24/7, security controls have to be largely invisible to the people running production.
“Whatever the controls you are putting, naturally, it has to be frictionless,” Singh says. Shop-floor teams are primarily concerned with uptime and production. If security makes their work unnecessarily difficult, adoption becomes harder.
This makes change management especially important. Plant employees are purpose-driven specialists, often with years of experience around specific machines and processes. Security teams therefore need to explain why a control is being introduced and how it benefits the plant rather than simply imposing another layer of compliance.
Resilience is more than prevention
Security also needs to be paired with recovery.
Singh argues that manufacturing organisations need both IT disaster recovery planning and a business continuity plan for the plant. The latter determines what happens when a line or system becomes unavailable — including whether production can continue manually, whether employees need to be reassigned, and how quickly operations need to recover.
Recovery objectives will differ depending on the business, customer commitments and production requirements. “How quickly you want to do it and then the data recovery time” determine how the recovery architecture should be designed.
AI adds another consideration: every new autonomous capability needs its own continuity plan. If an AI agent fails, is compromised or becomes unavailable, organisations need to know what takes over and how operations continue.
Security by design, not after deployment
For Singh, the answer is to bring cybersecurity into manufacturing technology decisions from the beginning.
At GEG, the security team follows a security-by-design approach, examining the architecture, integration points, dependencies, data, uptime requirements and the sensitivity of the environment before a new technology is introduced.
The approach is particularly important in OT, where a technology decision can remain embedded in the plant for years. Security, data protection and privacy therefore need to be considered alongside the technology itself.
As factories become more intelligent, the security perimeter is no longer limited to the IT network. It stretches from the machine and controller to the network, cloud, data, AI model and ultimately the physical action taken on the shop floor.
For Singh, building that chain securely requires a combination of segmentation, secure remote access, virtual patching, industrial DMZs, OT monitoring, governance and people awareness. The goal is not to slow down the intelligent factory but to ensure that greater connectivity does not turn its operational advantage into a new attack surface.