Industrial organisations are adopting synthetic intelligence for risk detection, community monitoring, and safety operations, however formal controls governing its use in operational expertise environments stay restricted, based on a brand new trade survey.
The State of AI in OT Cybersecurity 2026 report discovered that 87.7% of respondents are utilizing, evaluating, piloting, or planning to undertake AI for OT cybersecurity. Solely 7.9% have deployed the expertise throughout a number of safety features.
Nearly one-third of respondents, 30.8%, have deployed AI for at the very least one OT cybersecurity perform, whereas 37.1% are evaluating or piloting the expertise. An additional 19.9% count on to start implementation inside the subsequent 12 months, whereas 12.3% reported no present plans.
AI is already getting used for risk detection and alerting by 33.8% of respondents, community monitoring and anomaly detection by 31.5%, and safety operations centre help by 24.5%. Nevertheless, solely 15.6% reported having an enforced AI coverage particularly overlaying OT or industrial environments.
AI use centres on monitoring and detection
AI use is concentrated in monitoring and analyst-support features. Menace detection and alerting was the most typical software, adopted by community monitoring and anomaly detection.
One other 22.2% of respondents apply AI to incident response and triage. Though these makes use of usually help safety evaluation relatively than direct industrial management, they will have an effect on how alerts are categorized, prioritised, and investigated.
An incorrect classification, suppressed warning, or unsuitable response advice can affect choices involving gear availability, operational continuity, and security controls.
Use was decrease in vulnerability administration, cited by 19.5% of respondents, danger evaluation and prioritisation at 17.5%, and predictive upkeep or asset well being at 12.9%.
The report additionally discovered a spot between reported advantages and formal measurement. Whereas 32.4% of respondents stated AI had delivered a quantifiable or noticed enchancment, solely 8.6% might display a formally measured consequence.
Most respondents considered AI positively, with 69.9% saying its advantages in industrial cybersecurity outweigh its dangers. Nevertheless, solely 20.9% stated the advantages had been clearly higher than the dangers.
The most important group, 49%, took a extra certified place and recognized issues involving reliability, knowledge integrity, mannequin manipulation, and the necessity for human oversight.
“These findings point out a transparent curiosity in deploying AI methods, however OT and ICS organisations battle with how one can implement AI safely, successfully, and with operational management,” Jonathon Gordon, directing analyst at Takepoint Analysis, stated.
Broader deployment stays unusual. Whereas 45.7% of respondents have formally evaluated, piloted, or deployed agentic AI for an OT cybersecurity perform, solely 21.2% have moved the expertise into an lively pilot, proof of idea, or manufacturing surroundings.
Of that group, 16.2% are conducting pilots or proofs of idea. Manufacturing deployment stands at 5%.
Agentic AI refers to methods that may take actions with out step-by-step human instruction. In OT environments, the extent of authority granted to those methods determines whether or not they stay advisory instruments or can affect actions affecting industrial processes.
Though 78.7% reported some type of human oversight for relevant AI-driven choices, most relied on casual practices. Among the many 183 respondents for whom the query utilized, 55.7% had oversight that was not formally codified, whereas solely 23% had a documented and enforced human-in-the-loop protocol.
An additional 18.6% reported no outlined protocol, whereas 2.7% stated AI-driven choices had been largely automated.
Documented oversight was intently related to consequence mapping. Among the many 42 respondents with an enforced human-in-the-loop protocol, 90.5% had accomplished at the very least some evaluation of the AI-driven choices that might have an effect on bodily processes.
Implementation challenges had been concentrated in knowledge and present infrastructure. Information high quality, availability, and labelling was essentially the most generally cited barrier at 45.4%, adopted by integration with legacy OT methods at 42.4% and reliability issues in safety-critical environments at 38.7%.
AI failures carry bodily dangers
Issues additionally prolong to the bodily penalties of assaults in opposition to AI methods or operational knowledge. Round 20.5% of respondents categorized such assaults as a high operational danger, whereas 43.4% described them as an rising precedence.
General, 87.7% acknowledged at the very least some chance that an AI-related cyberattack might contribute to downtime, gear harm, or a security occasion. Concern was larger amongst organisations already utilizing AI, with 79.2% putting the difficulty in one of many two highest concern classes, in contrast with 37.3% of non-users.
The report stated the chance isn’t restricted to autonomous methods. A compromised AI software might suppress an alert, misclassify an incident, suggest an inappropriate response, or corrupt info utilized in an operational resolution.
Nozomi Networks CEO Edgard Capdeveille stated attackers are additionally utilizing AI to help cyber operations. “Adversaries are more and more utilizing AI to enhance their assaults, heightening the velocity and class of threats,” Capdeveille stated.
He stated industrial defenders are introducing AI into their safety operations in response. The survey, nevertheless, discovered that technical safeguards round these instruments stay restricted.
Simply over one-third of respondents, 35.1%, stated they’ve controls designed to guard AI instruments and fashions in opposition to manipulation, adversarial inputs, or provide chain compromise.
Solely 7.6% stated these safeguards had been examined sufficiently to help excessive confidence, whereas 27.5% had launched controls that weren’t but totally validated.
The report recognized coaching and operational knowledge, mannequin behaviour, manipulated inputs, third-party elements, and AI-generated outputs as areas requiring assurance. These dangers prolong past typical community and endpoint safety as a result of altered inputs or outputs can have an effect on how safety groups interpret exercise inside an OT surroundings.
BlastWave CEO and co-founder Tom Sego stated organisations ought to assume attackers have entry to the identical expertise. “Embrace AI for the whole lot it may possibly do in OT, however deploy it behind a fringe that assumes the adversary has AI too,” Sego stated.
OT-specific governance stays restricted
OT-specific governance insurance policies stay unusual. Solely 15.6% of respondents reported having an enforced coverage governing using AI in operational environments.
One other 34.8% had been creating an OT-specific AI coverage, bringing the share with an enforced coverage or one in growth to 50.3%.
An additional 30.8% relied on normal IT or cybersecurity insurance policies. The report stated such insurance policies might not account for AI-related operational, security, and physical-process dangers in industrial environments.
Governance exercise was extra widespread amongst organisations that had already deployed AI. Amongst respondents utilizing AI for at the very least one OT cybersecurity perform, 81.7% had both enforced an OT-specific coverage or had been creating one.
A written coverage didn’t at all times correspond with examined controls. Of the 47 organisations with an enforced OT-specific AI coverage, solely 17 had examined the safeguards defending their AI instruments and fashions.
The report additionally discovered that solely 11.9% had formally recognized and reviewed AI-driven choices that might immediately have an effect on bodily processes, security methods, or operational continuity. One other 25.2% had assessed chosen methods.
Consequence mapping hyperlinks an AI-generated output to the operational resolution it informs, the motion that will comply with, and the management accessible to cease an unsafe final result.
Among the many 176 respondents for whom consequence mapping utilized, 63.6% had accomplished at the very least some type of evaluation. Nevertheless, just one in 5 had established a proper, reviewed course of.
All 36 respondents with a formally mapped and reviewed course of additionally reported having an enforced OT-specific AI coverage. The report stated this affiliation signifies that formal governance and consequence mapping are likely to develop collectively.
Greater than half of respondents, 54.3%, had been monitoring or making ready for AI-related laws, frameworks, or sector steering, however solely 17.9% had established a proper initiative.
“Over the following six to 12 months, the organisations that progress furthest are more likely to be those who develop AI use with out granting it extra authority than their controls, proof, and working fashions can help,” Gordon stated.
(Photograph by Homa Home equipment)
See additionally: Anthropic and Nozomi deliver AI vulnerability analysis to OT safety
Wish to study extra in regards to the IoT from trade leaders? Try IoT Tech Expo going down in Amsterdam, California, and London. The great occasion is a part of TechEx and is co-located with different main expertise occasions together with AI & Massive Information Expo and the Cyber Safety Expo. Click on right here for extra info.
IoT Information is powered by TechForge Media. Discover different upcoming enterprise expertise occasions and webinars right here.