Introduction
In November 2025 IDC produced the whitepaper How AIoT Is Reshaping Industrial Efficiency, Security, and Decision-Making[1]. There are some really interesting findings from the paper but I wanted to focus in on two of the themes in particular as they have high resonance in the industries that I work with most. First the perennial AIoT use case, predictive maintenance, which has been a key challenge for as long as I’ve been working with the energy industry. And security as the explosion of distributed energy resources, and smart home products has massively increased the potential attack surface of our critical national infrastructure.
Predictive Maintenance
It should be somewhat unsurprising that predictive maintenance tops the list of AIoT use cases identified in the report. Unplanned downtime is the single greatest threat to productivity and therefore profitability in the sector.
Predictive maintenance has been a mainstay of similar reviews for several years, indicating that the requirement persists. There are number of reasons for this, not limited to the fundamentals of data quality and data availability which continue to challenge the most businesses, particularly at the interface between IT and OT. Additionally, predictive maintenance is not a whole cloth replacement for regular planned maintenance or condition-based maintenance of high-cost assets with long replacement lead times. Where it is particularly useful is in the planning process, to ensure that when an engineer is on-site they have the correct parts available and on hand, improving the first fix rate and reducing the number of revisits required. This reduces downtime for the equipment and makes the overall maintenance process more efficient.
Supply chain implications
Even in processes where predictive maintenance may not seem appropriate due to cost of early replacement of high capital assets it can be a useful companion process. In many scenarios these high replacement cost parts also have long delivery lead times due to low manufacturer inventory. In this case, predictive maintenance algorithms can be used to plan the forward purchasing of these items to more closely align the capital cost requirement with the maintenance outage required. This enables companies to free up capital by avoiding overstocking and ensures that parts are ordered in time for their likely utilisation.
The role of Generative AI
Operations and maintenance (O&M) is an area where generative AI can be deployed to great effect and with a clear line between the deployment of the technology and improved productivity.
Organisations are replete with technical documents, maintenance records and field engineers’ reports. Using retrieval augmented generation (RAG) companies can interrogate this rich source of institutional knowledge that was previously hard to leverage. One practical way this can manifest is in the upskilling of O&M engineers by presenting this information to them at the point of use. This enables even an engineer that is relatively new to the organisation to take advantage of the stored knowledge about the equipment they are about to interact with. Combined with data from condition-based monitoring and/or concurrent observation, this solution can suggest practical interventions based on the precise history of the equipment in question.
Agents in Action
Another new trend across industry, tied to generative AI, is the development of autonomous AI “Agents” to proactively monitor equipment and then act as both an alerting mechanism and decision support tool for the O&M engineer. This agentic approach effectively creates a team of junior engineers that focus on specific pieces of equipment and then “report back” to the human engineer for oversight and physical interventions. While this technology is still largely in the experimental stage it has the potential to alleviate ongoing resource constraints by allowing a single engineer to proactively monitor and assess many more resources than they could do alone. By the implementation of carefully considered feedback and control loops these agents can be enabled to take more decisions automatically such as the creation of work orders, parts ordering or event direct adjustment of digitally controlled parameters.
In Summary
Predictive maintenance exists along a continuum with reactive, fix-for-fail, at one end, through scheduled and condition-based maintenance all the way to fully predictive, just-in-time schemes at the other. I a mixed approach of scheduled, condition-based and predictive, enhanced by RAG, is likely to deliver the most cost-effective solution overall but companies should look to the future and plan for agentic workloads to play an ever-larger role. To justify investment and prove value companies should prioritize use cases where there is a clear ROI in the form of increased productivity, cost management or supply chain capital optimisation.
Emergent risks of intelligent “things”
The challenges of scale
By the end of 2025 it is estimated that there are around 20 billion IoT devices [2] and that number is set to double by 2034. This rapid explosion in device numbers causes two key risks.
First, the attack surface is expanding at an incredible rate, with every connected device presenting a potential point of entry for bad actors.
Secondly these numbers are made up of an increasing diversity of devices, often abandoned or orphaned by manufacturers and poorly understood by consumers in terms of firmware updates to patch out critical vulnerabilities.
Companies deploying IoT devices must ensure that network perimeter security is tight and that asset registers are correctly completed, maintained and monitored to ensure the latest security patches are applied in good order. Poor firmware management has been responsible for an increasing number of cybersecurity incidents in 2025 [3]
New exploits
Increasing use of AI in IoT systems also presents new attack vectors for malicious actors. AIoT devices increasingly rely on onboard or cloud‑linked AI models for decision‑making. Attackers can manipulate these models by corrupting training or operational data, causing devices to behave incorrectly or dangerously.
Data poisoning can distort an AI system’s perception of the environment, degrading accuracy or forcing malfunctioning behaviours which, in OT coupled systems, have real world impacts.
AI‑targeted manipulation represents a shift from exploiting devices to corrupting intelligence, a newer and more complex threat surface. AI models enable sophisticated pattern recognition, which also enables attackers to extract sensitive information.
Inference attacks can reconstruct training data or uncover private patterns from AIoT device outputs. As AIoT devices continuously collect ambient data (video, audio, biometrics), inference‑driven privacy leaks become significantly more damaging.
AI integration enables botnets to become more adaptive, stealthy, and decentralized. Reports forecast autonomous botnets capable of self‑learning and coordinated swarm‑like attacks. Real‑world IoT compromise activity is rising sharply, with examples of 13,000‑device botnets participating in record‑breaking DDoS attacks.
In Summary
While AIoT promises huge benefits, there are significant new risks presented by the integration of AI into the IoT domain. The good news is that AI itself can be used to help manage and mitigate these risks. SAS Information Governance Catalogue, SAS Model Manager and SAS AI Governance explorer make it easier for teams to monitor and control deployed models and IoT endpoints to intervene when necessary.
The Takeaway
The use of Industrial AI has the potential to bridge the gap between Operational Technology and Information Technology. Companies that embrace this change will have a strategic advantage over those who do not. But risks do exist and these should be proactively managed to ensure that AI is delivered in a trustworthy manner.
These are not just technological challenges though. Writing from experience, the greatest barrier to any successful AI implementation is not the software. It is operational readiness, if end users do not trust the software, or find it hard to use then adoption will be poor and ROI will be hard to achieve.
SAS is able to draw on a five decade heritage of innovation, supporting customers across manufacturing, heavy industry and energy to deliver best in class Industrial AI solutions that create that bridge and deliver on the promise of AI to drive efficiency, innovation and return on investment inside a well-established trust and security framework.
With trusted solutions to fit every type of user, from engineers to data scientists, no-code to yes-code, SAS enables everyone to be part of the AI revolution.
Find out more about Industrial AI at SAS by downloading the ebook Industrial AI | SAS.
[1] How AIoT Is Reshaping Industrial Efficiency, Security, and Decision-Making
[2] How Many IoT Devices Are There [2026 Statistics]
[3] IoT Security Breaches Report and Industry Insights 2025
