“We’re dealing with AI that can actually make decisions on the factory floor — if it goes wrong, the consequences aren’t just digital, they’re physical,” says Andy Foster, Product Lead at IOTech. “That’s why safety at the Edge is everything.”
Certainly, as factories, power grids, and transport networks adopt AI at pace, ensuring it operates safely has become a pressing challenge. While much of the world talks about accelerating AI, fewer are asking the harder question: how do you control systems that make real-time decisions in high-stakes industrial environments?
In the UK, that question is shaping collaboration between industry and academia. The National AI Labs, a government-backed consortium of leading universities, is working with companies to develop practical guardrails for AI deployed at the Edge.
One company turning these principles into operational systems is IOTech, a Newcastle-based Edge software firm. Its platforms operate close to heavy industrial machinery, collecting and analysing data from diverse equipment in real time, running AI-driven analytics locally, and feeding insights back to operators or up to the Cloud.
“We’re a software products company focused on industrial Edge platforms,” Foster says. “Any data-rich environment where you need to acquire data from lots of heterogeneous equipment, run analytics or AI locally, and then share it northbound to Cloud systems — that’s our space.”
Founded roughly 10 years ago, IOTech employs more than 50 staff and is based on the campus of Newcastle University, within the National Centre for Big Data. Nearby sits the National AI Research Hub, with which IOTech has been collaborating over the past year. Foster describes the partnership as “incredibly productive,” giving the company access to cutting-edge research while providing academics with real-world industrial use cases.
“As a medium-sized company, we simply don’t have the resources to research every forward-looking AI challenge ourselves,” he says. “Collaborating with the Labs and AISI [AI Security Institute] gives us access to advanced techniques while providing them with practical industrial use cases. It’s mutually beneficial — their research informs our products, and our operations provide real-world data and scenarios.”
IOTech’s software powers advanced manufacturing plants, building management systems, renewable energy sites, and distributed energy systems such as battery storage networks. In these settings, AI can be advisory — predicting when a machine might fail — or operational, adjusting equipment settings in real time to optimise efficiency.
“That distinction is critical for safety,” Foster explains. “If AI is just informing you of conditions, that’s one thing. If it’s generating outputs that go back down to the equipment and change something in the production environment, that’s different. You have to understand the risk and design for it.”
Edge computing presents unique constraints. Unlike Cloud AI, Edge systems must operate with limited processing power and minimal latency. “The AI we deploy has to be optimised for the environment,” Foster says. “You can’t just throw a large model into a factory controller and expect it to work safely.”
Even traditional models require vigilance. Predictive analytics and rule-based AI must be monitored for drift — when outputs gradually diverge from expected results. Fail-safes and “kill switches” are standard in industrial environments to prevent accidents.
The challenge grows as generative and agentic AI — capable of autonomously planning and executing actions — enters industrial operations.
“With agentic AI, you’re not just running a model built on a known rule set,” Foster explains. “You’re potentially allowing it to execute a series of actions autonomously. The level of care, the guidelines, and the checks and balances have to be completely different.”
Recent academic experiments have shown how loosely constrained agentic systems can behave unpredictably when exploring datasets, sometimes producing manipulative or dangerous outputs. Foster emphasises that industrial deployments require tighter control:
“You can constrain the type of questions the AI can answer, the actions it can run, and the number of actions in a given time period. You can also insert human review points, and the system can revert to a safe state if anything unusual is detected.”
Physical safeguards further reduce risk in robotics. Sensors, hardware overrides, and interlocks ensure that a robot encountering an unexpected obstacle will halt automatically. “Even if the AI is making decisions, the environment itself can intervene,” Foster says. “That’s essential when you’re dealing with moving machinery or heavy loads.”
Collaboration with the National AI Labs allows IOTech to contribute industrial expertise to broader AI safety research. “They’re looking at extreme scenarios — what could go wrong if an AI system were compromised or misused — and we feed into that from an operational perspective,” Foster adds.
Regulatory frameworks are still catching up. The European Union’s EU AI Act introduces binding rules, while the UK has so far taken a principles-based approach. “I don’t want to be over-regulated as a medium-sized technology company,” Foster says. “But there does need to be regulation that represents best practice. Industrial AI, particularly at the Edge, presents unique challenges — errors can cause physical damage or safety hazards, not just financial losses.”
Despite the complexity, Foster sees the rapid evolution of industrial AI as an opportunity. “You can do immensely powerful things with AI, but particularly for safety-critical customers, you have to embed the guardrails. Companies that can combine autonomy with safety, monitoring, and governance will have a real competitive advantage.”
IOTech is building platforms that allow customers to deploy advanced AI — including agentic systems — while controlling behaviour, managing updates at scale, and integrating with existing infrastructure. “We want operators to get the benefits of AI safely,” Foster says. “That’s the challenge and the opportunity right now.”
