The modern factory is becoming more connected, intelligent and dependent on real-time decisions. Industrial machines, operational technology (OT), Industrial Internet of Things (IIoT) sensors and enterprise systems are increasingly interconnected, enabling manufacturers to improve productivity, anticipate equipment failures and optimise production. But this convergence is also changing the nature of cybersecurity risk.
As artificial intelligence moves from analysing manufacturing data to influencing operational decisions, the consequences of a cyber incident extend beyond compromised information or disrupted IT systems. A manipulated data feed, compromised industrial endpoint or corrupted AI model could affect production quality, equipment reliability and, in high-impact environments, worker safety.
For manufacturers, therefore, cybersecurity is no longer simply about keeping attackers out. It is about ensuring that factories can continue operating safely, recover from disruption and trust the systems guiding their decisions.
“Cyber resilience is not a technical checkbox owned by the security team — it is a business continuity and enterprise risk issue that belongs in the boardroom,” says Pradipta Patro, Chief Information Security Officer and Vice President, Polycab India Limited.
Production reliability, brand reputation, regulatory compliance, worker safety and customer trust all depend on an organisation’s ability to withstand and recover from cyber disruption, Patro argues. Forward-looking manufacturers must consequently integrate cyber resilience into operational strategy, capital planning and risk governance, with the same rigour applied to quality, safety and supply-chain continuity.
When AI becomes part of the attack surface
AI promises substantial gains for manufacturing, from predictive maintenance to autonomous process optimisation. However, as algorithms begin to influence and potentially make operational decisions, manufacturers must account for risks that conventional security controls may not adequately address.
Patro identifies three priorities before AI is scaled into operational environments: data integrity, model governance and decision-making oversight.
“AI is only as trustworthy as its data,” he says.
Corrupted or deliberately poisoned inputs can lead an AI system to generate incorrect recommendations or potentially unsafe decisions without an obvious indication that something has gone wrong. Manufacturers must therefore validate and continuously monitor the data pipelines feeding their models.
The models themselves also require protection. Patro emphasises least-privilege access, authorised changes, logging and reversibility to prevent AI systems from becoming an indirect route into critical operations.
The third requirement is oversight. In environments where production quality and safety are at stake, an unexplained AI recommendation cannot automatically be treated as reliable.
Manufacturers need explainable AI and meaningful human oversight for high-impact decisions. The objective is to ensure that operational teams can understand, challenge and intervene when an AI-driven recommendation appears inconsistent with actual plant conditions.
Visibility is the foundation of OT security
The expanding number of connected machines and industrial endpoints presents another challenge: maintaining an accurate understanding of what is connected to the factory environment and how those assets communicate.
“You cannot protect what you cannot see,” Patro points out.
Resilience begins with complete, real-time visibility across the industrial environment, from programmable logic controllers (PLCs) and supervisory control and data acquisition (SCADA) nodes to engineering workstations and vendor connections.
This visibility provides the foundation for determining which assets require protection, identifying unexpected activity and understanding how a compromise could spread.
Network segmentation is equally important. Separating IT and OT environments into distinct security zones helps contain a compromise rather than allowing it to propagate across connected systems. Least-privilege access must also apply to users and devices, while continuous monitoring should identify anomalies using capabilities suited to industrial environments.
The challenge is not simply to account for every connected device. It is to establish disciplined visibility, containment and control across an increasingly complex attack surface.
Security must be designed into the smart factory
For Patro, securing a smart factory cannot be treated as a post-deployment exercise.
“Security in a smart factory cannot be retrofitted, it must be engineered from the first blueprint,” he says.
Four practices are particularly important. The first is network segmentation, which creates defensible boundaries between IT and OT systems. The second is strong identity and least-privilege access control for every user, device and connection.
The third is secure-by-design architecture, incorporating redundancy, offline backups and safe-failure states. These measures help ensure that an incident does not automatically become a prolonged operational disruption.
The fourth is continuous visibility through real-time asset intelligence and monitoring.
These foundations are particularly important as manufacturers modernise legacy infrastructure and connect previously isolated systems to enterprise networks, remote service providers and cloud environments.
Embedding security during architecture and deployment also reduces the need for expensive remediation later. More importantly, it makes resilience a characteristic of the factory’s design rather than a collection of controls added after vulnerabilities emerge.
The next challenge: Trusting autonomous decisions
Looking three years ahead, Patro identifies trust in autonomous decision-making as the biggest cybersecurity challenge facing AI-driven manufacturing.
As AI takes on a greater role in operations, attackers may increasingly seek to manipulate the intelligence guiding those operations rather than simply steal data or disrupt networks.
Patro points to data poisoning, adversarial inputs and model tampering as threats that could subtly influence AI outputs and steer machines towards unsafe or costly decisions.
“Traditional defences protect networks, not models, and few can detect when an algorithm has been subtly compromised,” he says.
This creates a different kind of security challenge. A factory may have network controls in place, and its machines may remain operational, yet the decisions being generated by an AI system could still be unreliable if the underlying data or model has been compromised.
Preparedness, therefore, requires manufacturers to secure data pipelines, validate models continuously and preserve meaningful human oversight over critical decisions.
For CIOs and CISOs, this means extending the security conversation beyond conventional infrastructure protection to encompass the integrity of the intelligence used in production.
The broader implication is that manufacturing cyber resilience must evolve alongside factory automation. Connected assets need visibility, industrial networks need defensible boundaries, and AI systems need controls that protect both their inputs and their decisions.
The factories that succeed will not simply be those that deploy AI most aggressively. They will be those that can demonstrate why their AI systems are trustworthy, intervene when decisions become questionable and recover when technology fails.
As Patro puts it, “The factories that thrive will be those that learn to trust their AI, because they secured it first.”