17 February 2025
Would AI flaws cause obstacles to the use of the systems in the OT environment, and how can any flaws get fixed? Suzanne Gill information.
Today artificial intelligence ( AI ) and machine learning ( ML) is transforming industrial processes through its diverse capabilities – such as predictive maintenance, process optimisation, quality assurance, and supply chain management. For instance, Siemens is using AI in its own factories to identify products malfunctions and simplify maintenance procedures while General Electric is using predictive servicing solutions that are powered by AI in power plants to identify turbine issues early to save costly repairs and functional disruptions.
AI/ML is also becoming a useful tool for automating and improving security procedures in the workplace. It can detect attacks and anomalies in method and system behavior, drive threat intelligence, generate threat-hunting hypotheses, prioritise vulnerabilities, analyse source code, estimate cyber risk scores, improve training and awareness, and implement incident response. Nozomi Networks is, for example utilising AI/ML algorithms to assess and relate business community data, including alerts, vulnerabilities, communications, and process telemetry to detect suspicious behaviours and complex security or configuration issues. According to a 2024 IBM report, organizations that automate cybersecurity saved an average of$ 1.88 million and cut breach containment by 108 days.
Overlooking security
However, according to Anton Shipulin, Industrial Cybersecurity Evangelist at Nozomi Networks, the security of AI/ML technology is often overlooked during its adoption within operational technology ( OT ) environments. He claimed that “adoption is largely influenced by digital transformation initiatives and the extensive offerings of industrial automation vendors.” ” In reality, beyond traditional cyberattacks on AI/ML infrastructure, AI/ML-based systems may be vulnerable to specific attack techniques such as evasion attacks, extraction of sensitive training data, and data poisoning attacks, which can impact operations and production processes”.
Security threats must be carefully taken into account as part of the adoption process, for this reason. Anton pointed to MITRE ATLAS ( Adversarial Threat Landscape for Artificial-Intelligence Systems ) as a valuable framework for addressing such challenges. Based on real-world attacks and demonstrations by AI red teams and security experts, he said,” MITRE ATLAS is a living knowledge base of adversary tactics and techniques targeting AI-enabled systems.” To address these techniques and counteract vulnerabilities, the framework also includes mitigations.
As AI/ML technology becomes a key component of on-premises or cloud systems as well as supply chains, Anton continued, it should be incorporated into a wider cyber risk management program. These initiatives are frequently motivated by local cybersecurity laws. For critical infrastructure companies in the EU, these requirements are defined by the NIS2 Directive, which emphasises vulnerability management, supply chain security, threat detection, incident response, and reporting. In this sense, the European Union’s NIS2 Directive is a positive step toward improving the security of AI adoption in OT environments”.
A valuable framework
Michael Schrapp, Global Head of Industrial AI Innovations, Siemens AG, believes that the NIS2 Directive provides a valuable framework to address vulnerability in the OT environment, where critical industrial control systems and critical infrastructure requires high levels of reliability, security, and safety. He acknowledges that even though AI has significant advantages in terms of optimization and automation, AI can also introduce new security risks that could compromise these mission-critical systems. The risk of manipulating or using AI models, which could have unintended and potentially dangerous effects on industries, is a top concern. The complexity and opaqueness of some AI algorithms also make it difficult to fully validate their decision-making processes, which is crucial for safety-critical applications”, he said.
NIS2 emphasises the importance of robust risk management, incident response, and the implementation of appropriate security measures. It is possible to reduce the risks associated with AI technology by aligning the adoption of AI in the OT environment with the NIS2 requirements, according to Michael. He said:” This involves implementing rigorous security controls, conducting thorough risk assessments, and ensuring AI systems are designed, deployed, and monitored just as carefully as other critical components in the OT environment” . ,
Michael believes that AI-powered predictive maintenance will enable industrial operators to better anticipate equipment failures and schedule maintenance proactively in the near future. It is possible to determine when components are most likely to malfunction or degrade by using advanced AI models to analyze sensor data and historical maintenance records.  ,
” Additionally, I see AI being leveraged to optimise industrial processes and workflows”, he said. The rise of generative AI applications is accelerating the impact of AI in industrial settings by automating decision-making and dynamically altering parameters, as demonstrated by AI algorithms ‘ ability to identify patterns, anomalies, and opportunities for improvement in real-time production data. In this way AI-powered process optimisation will help industrial facilities run more efficiently, reduce waste, and increase yields.
” In the longer-term, I see AI being used for autonomous control of complex systems. Robots and cobots that are AI-driven will seamlessly work alongside humans, increasing safety and flexibility on the factory floor. ”  ,
Maintenance and control engineers will need to develop their skillsets and adapt to new ways of working, according to Michael in his analysis of the future. Maintenance engineers will need to become experts at analyzing sensor data, interpreting AI-generated insights, and utilizing that knowledge to improve maintenance schedules and workflows. They will also be essential in ensuring that the industrial assets they monitor are properly trained, calibrated, and integrated with the AI models.  ,
In the future, maintenance engineers will be able to communicate directly with machines, asking about issues or the status that they may have. This will help identify issues as well as provide solutions. The integration of AI will require a thorough understanding of how these systems interact with and influence the control logic for control engineers. To ensure that AI’s autonomous decision-making is consistent with the overall control strategy and safety protocols, control engineers will need to work closely with AI developers. They will also be responsible for ensuring the integrity of the control architecture, troubleshooting, and monitoring the performance of AI-based control systems.
Careful evaluation
The use of AI in OT environments requires careful analysis of the risks and capabilities of each specific application, according to Oakley Cox, Director of Product at Darktrace. ” While AI has proven effective in OT cybersecurity applications, particularly for anomaly detection and threat response, not all AI types are suitable for these critical environments”, he warned. ” For example, Generative AI presents significant challenges around data sovereignty, accuracy, and transparency that make it unsuitable for OT systems where reliability and predictability are paramount”.
In the near future, Oakley believes that improving cybersecurity posture and risk management will be the most important impact. By automatically identifying assets, identifying risks, and identifying anomalies, he said,” AI-powered solutions are already changing how organizations protect their OT environments.” This is particularly important as ransomware and other threats increasingly target OT systems, where attacks can result in not only financial losses but also operational downtime and outages.
Looking to the longer-term Oakley believes that AI will drive deeper IT/OT convergence, enabling more sophisticated integration between business systems and operational technologies to create opportunities for dramatic improvements in production efficiency, maintenance, and scaling. ” Organisations will move beyond using AI just for security to leveraging it for predictive maintenance, process optimisation, and automated decision-making across their cyber-physical systems”, he said. The shift from traditional rule-based methods to AI-driven approaches that combine multiple machine learning techniques is where the realisation of these benefits lies. These systems will increasingly use pattern analysis, probabilistic modelling, and relationship analysis to understand the complete’ pattern of life’ of an organisation’s assets, enabling both better security and operational improvements. However, success will require organizations to shift from conceiving IT and OT in silos and adopt a more holistic approach to their cyber-physical infrastructure.