
How do you create secure OT architecture for AI? For control engineers and system integrators, Skkynet CEO Gary Tillery’s advice starts with network segmentation, usable legacy data and security designed in from the start.
The request sounds simple: Get plant-floor data into an AI application. But the underlying engineering question is much harder. What else are you connecting when you make that connection?
For Gary Tillery, CEO of industrial connectivity company Skkynet, that question belongs at the beginning of a project, not after the analytics platform is selected or the network connection is working. Manufacturers need access to operational data, but the architecture used to deliver it deserves as much attention as the application consuming it.
Speaking on the Ctrl+Alt+Mfg podcast, Tillery offered a blunt response to plant teams worried about exposing their control systems: “Be nervous.”
That concern, he argued, should drive better engineering rather than prevent data sharing. His recommendations center on understanding the operational problem, scrutinizing connection paths and resisting architectures that make everything accessible through one place.
Start with the problem, not the product
Tillery sees system integrators as essential to that work because their contribution extends beyond installing software or connecting controllers to operator interfaces.
“The system integrator is doing more than just integrating systems,” he said.
Customers bring operational problems and business objectives, and vendors supply technology. Integrators translate between them, applying domain knowledge, troubleshooting skills and architectural judgment to determine what a workable solution requires.
That distinction matters when defining project scope. Moving data to a cloud application may call for a different architecture than exchanging information between IT and OT systems. Treating both as generic connectivity assignments overlooks the network conditions and security concerns that shape the design.
Tillery pointed to root-cause questioning, including the five whys, as a way to understand the customer’s actual need before choosing tools. For engineering teams, the practical lesson is to engage integrators in problem definition and architecture, versus simply looking at implementation.
“They can break the problem down into a very root-cause way,” Tillery said.
Make existing data useful
Before pursuing new analytics, teams also need to determine whether existing operational data can reach the intended application in a usable form.
“There’s not a lack of data,” Tillery said. “It’s in the wrong format, or it’s the poor quality, or there’s other issues.”
He described Skkynet’s approach as a data backbone connecting legacy sources with newer applications. That includes translating data from established industrial protocols such as Modbus for use through newer connectivity options, rather than requiring individually customized mappings.
The broader engineering consideration is how much integration work a proposed solution leaves to the project team. It’s important to evaluate the path from the existing device or system to the consuming application, including translation and data quality.
Making more data available is useful only if the resulting information can support the intended task. Connectivity and usability belong in the same conversation.
Scrutinize inbound access
When discussing common security mistakes, Tillery singled out opening inbound ports to obtain plant data.
“You have to remove the attack surface, which is an inbound port,” he said.
He described Cogent DataHub’s outbound-initiated connection approach as a way to exchange data without opening inbound ports into the protected network. He also described the data exchange as bidirectional. Connection initiation and permitted data direction are separate architectural questions.
For control engineers evaluating such an approach, that distinction deserves explicit discussion. An outbound-initiated connection should not be confused with a physically unidirectional data path, nor should eliminating an inbound port be interpreted as eliminating every security risk.
Tillery’s central recommendation was to examine the design from an attacker’s perspective. Consider not only how information reaches its destination, but what an intrusion elsewhere in the environment could expose.
“You should be looking at it from the attacker’s side,” he said.
Avoid the monolithic namespace shortcut
Tillery cautioned against building a unified namespace, or UNS, around a single monolithic location through which every network exchanges data.
His concern combines security and communications architecture. Concentrating access can concentrate exposure, while repeatedly moving information to one location and back can create unnecessary traffic.
“You don’t want every network, if you have a large network system, to go to that one location,” Tillery said.
Skkynet calls its distributed approach a fractal UNS. Beyond the terminology, Tillery’s recommendation is to segment the architecture rather than make one central system the default destination for everything.
The design question is therefore not simply whether the enterprise can access plant information; it is whether the arrangement preserves appropriate boundaries while delivering that information where needed.
Tillery also emphasized the handoff to plant personnel. An architecture must remain manageable and expandable after the integrator leaves, since the end user will operate it without the project team continuously present.
Demand evidence, including security by design
While a roadmap is important, Tillery stressed that product evaluation should test what vendors realistically can deliver today. Discussing live demonstrations, he emphasized showing working software, explaining implementation effort and distinguishing available capabilities from future promises.
That scrutiny should extend to development practices. Tillery said Skkynet moved toward quarterly releases to deliver functionality and security updates in smaller increments, so customers can decide when to use them in production.
He also described security review as a design activity conducted before code is written or features are added.
“From the very beginning, everything is challenged on the design first before there’s any code written,” he said.
For engineering teams, the takeaway is to ask vendors to demonstrate both the connection and the thinking behind it.
“You have to be able to share the data,” Tillery said.
The Ctrl+Alt+Mfg Podcast
Make sure to check out other episodes of the Ctrl+Alt+Mfg podcast, where hosts Gary Cohen and Stephanie Neil discuss a range of digital transformation insights. The last five episodes are listed below:
Ep. 18: Audacious Goals for Infrastructure Security, with Danielle Jablanski of STV
Ep. 19: The Future of System Integration, with Daren Dieleman of Interstates
Ep. 21: The Business Behind Automation, with Jeremy Anderson of Process and Data Automation
Ep. 22: Scaling the Modern System Integrator, with John Sullivan of DMC