Why OT and AI belong in the same conversation
AI is arriving in industrial environments through ordinary software updates. The security conversation needs to catch up.
Picture a maintenance planner at a mid-sized manufacturer. Their scheduling software has just updated, and it now offers an assistant that reads equipment logs and suggests when to take a line down for servicing. Nobody held a meeting about it. Nobody asked the security team. It simply appeared, because that is how most AI reaches a business today.
That small moment is where operational technology security and AI security stop being separate subjects.
Two teams, one problem
In most organisations, OT security and AI governance live in different places. OT sits with engineering and operations, focused on availability and safety. AI sits with IT, data or a newly formed working group, focused on policy and productivity. Each has its own vocabulary, its own risk register and its own budget.
The trouble is that the assistant in the scheduling tool does not respect that boundary. It reads data from the plant, it runs in a tool the business relies on, and it may send information somewhere outside the site. Both teams own part of the risk, and neither owns all of it.
The questions that matter
You do not need a new framework to start. A handful of plain questions will tell you most of what you need to know:
- What can the AI feature see, and does that include data from control systems or plant networks?
- Where does that data go, and is it kept or used to train anything?
- Can the feature only suggest, or can it act, for example by changing a setting or a schedule?
- Who approved it being switched on, and who would notice if it behaved strangely?
If the answer to the third question is that it can act, the conversation changes. A suggestion that a person reviews is a very different risk from an automated change in an environment where downtime costs money and mistakes can affect safety.
The useful line is not between OT and AI. It is between what only advises and what can act.
Start with an inventory
Most organisations already know, roughly, which systems run their plant. Far fewer know which of those systems now include AI features, because vendors add them in routine releases. A short exercise to list the industrial and business tools in use, and to check which have gained AI capabilities in the last year, is often the most valuable first step. It is unglamorous, and it is where good decisions begin.
What this site will cover
These notes will look at both sides and, most of all, the overlap: how to assess an AI feature before it touches plant data, what regulators are starting to expect, and how to keep the conversation practical for people who are not security specialists. If that is a problem you are wrestling with, you are in the right place.