In late 2023, an industrial robot at a vegetable processing plant in South Korea crushed a worker to death. Its vision system had misidentified the man as a box of produce. This was not science fiction or some distant theoretical risk. It was a working factory, a routine inspection, and an AI system doing exactly what it had been built to do - just not what anyone intended.
That incident was a preview, not an exception. Factories today are deploying AI that does far more than reading a dashboard. It acts. It adjusts set points, reroutes material, schedules maintenance, and increasingly makes decisions on the shop floor without a human approving every step. As that autonomy grows, one question keeps getting louder: who is actually in control?
The Promise is real - and so is the Gap
The upside is enormous. AI in manufacturing is already detecting equipment failures before they happen, cutting unplanned downtime, optimizing energy use, and improving safety in ways that were not possible a decade ago. McKinsey estimates that agentic AI - systems that plan and act on their own could add between $2.6 and $4.4 trillion to the global economy each year.
Yet the results on the ground tell a more sobering story. In Grant Thornton's 2026 AI Impact Survey, not single manufacturing respondent reported significant revenue uplift from AI when compared to 12% of executives across all other industries. The technology has clearly arrived in manufacturing. The returns, for most, have not. And the reason is rarely the algorithm. It is the absence of governance built for the realities of a factory.
The conversation is happening in the wrong rooms
Pick up almost any report on AI governance today and you will find it framed around banking, big tech, or generic ethics: bias in lending, transparency in hiring, fairness in customer service. These are important conversations but they are written for industries where a wrong AI decision usually costs money or reputation.
In asset-heavy industries such as aluminium, energy, chemicals, process manufacturing a wrong decision can cost a furnace, a shipment, an environmental fine, or a life. The stakes are physical, not just financial. Yet these industries are barely part of the mainstream governance discussion, and the frameworks being offered to them were designed for a very different world.
The numbers reflect this blind spot. Three out of four organizations now admit their governance has not kept pace with their AI adoption, according to Informatica's 2026 research. More worryingly, a 2026 enterprise survey found that 36% of executives have no formal plan for supervising AI agents, and 35% admit they could not immediately "pull the plug" on a rogue agent. In a factory, that is not a compliance gap. That is an operational hazard.
Also Read: Enhancing Agricultural Productivity Through AI-Powered Digital Tools
Four questions every industrial leader needs to answer
The governance challenge in heavy industry comes down to four hard questions. None are about the technology itself. All are about how we manage it.
When an AI agent makes the wrong call, who owns it?
A passive tool produces a recommendation a human accepts or rejects. An agentic system takes the action. So when an autonomous agent throttles the wrong unit or misreads a sensor, accountability becomes murky - the vendor, the IT team, the plant manager, or the operator who did not override it in time? Until accountability is named before deployment, every incident becomes a blame exercise after the fact.
How much autonomy is too much?
An AI agent optimizing a maintenance schedule is a very different risk from the one controlling a smelter's temperature. The discipline lies in matching autonomy to risk for each process - from systems that only advise, to systems that act within tight boundaries, to fully closed-loop operation rather than applying a single policy across the plant. Most manufacturers are still early on this curve, and a few have drawn these lines explicitly.
How do we govern AI sitting on decades-old systems?
This is the challenge unique to heavy industry. Most plants run on operational technology - PLCs, SCADA, DCS - installed 15 to 25 years ago and never designed to be supervised by adaptive AI. As IT and OT converge, AI ends up making decisions on infrastructure that predates the smartphone. Governing that intersection is a no-man's-land that traditional frameworks like COBIT or ITIL were never built to cover.
How do we earn the trust of the people on the floor?
AI on a shop floor works alongside operators, supervisors, and middle managers. If they do not trust it, they will override it, ignore it, or quietly work around it - and the investment dies. Only about a third of employees say their manager is an AI champion. Technology adoption in a factory is ultimately a human decision, made shift after shift, by people who need to believe the system has their back.
Also Read: How AI Servers Outperform Traditional Servers
What smarter governance looks like?
A governance framework for smart factories does not need to be a 200-page document. The organizations getting this right share a few principles. They assign clear ownership. Every AI agent has a named human accountable for its decisions, not a committee. They tier autonomy to risk, keeping safety-critical systems under tight human control. They build in a genuine, tested ability to stop an agent instantly. They monitor continuously rather than auditing once a year, because an agent's behavior can drift over time. And they bring OT and IT governance together instead of leaving the gap between them ungoverned.
Emerging standards such as ISO 42001 - an AI management system framework in the spirit of ISO 27001 are starting to help. But standards are scaffolding, not a substitute. Much of the real work is industry-specific, and heavy industry will have to do it for itself. There is also a people dimension no framework can skip. BCG's "10-20-70" rule is a useful reminder: successful AI transformation is roughly 10% algorithms, 20% technology and data, and 70% people and process. The plants that succeed will invest as heavily in training and trust as in the technology itself.
The path forward
The factories that win the next decade will not be the most automated. They will be the best governed. Automation is becoming a commodity; the ability to deploy autonomous AI safely, accountably, and at scale is what will separate leaders from laggards.
Smart factories are no longer a vision of the future - they are being built right now. The question is whether the governance around them can keep pace with the intelligence inside them. For asset-heavy industries, getting this right is not a regulatory box to tick. It is the difference between AI that creates lasting value and AI that becomes the next cautionary headline.
Smart factories need smarter governance. The sooner our industry treats that as a leadership priority rather than an IT afterthought, the better.
About the Author:
Ankur Bakliwal, A digital transformation and IT leader with over 22 years of experience across global enterprises. He Manager – Digital Solutions at Norsk Hydro, leading large-scale projects, service delivery, and digital initiatives across multiple teams. I am certified PMP, ITIL, and Six Sigma practitioner.