To combat the escalating threat to manufacturing systems, many security operations centers (SOCs) are implementing agentic AI to preemptively identify and mitigate potential vulnerabilities.
The State of Agentic Cybersecurity in Manufacturing
Lee Rossey, CTO and Co-Founder | SimSpace
While ransomware attacks on hospitals and data breaches at financial institutions tend to get more than their share of media attention, manufacturing remains the most attractive target to persistent threat actors (PTAs) seeking to cause disruption to critical infrastructure.
Manufacturers are uniquely vulnerable to cyberattacks by malicious actors. Even moderate disruption to production facilities can have lasting financial impacts, and the often-global nature of modern supply chains poses significant risk of secondary and tertiary impacts to thousands of organizations and businesses, and potentially millions of consumers.
According to Check Point Software Technologies, cyberattacks against manufacturers increased by 56% between 2024 and 2025[i], with legacy operational technology (OT) systems being a crucial attack vector. For decades, OT security was overlooked on the assumption that air-gapped industrial technologies such as programmable logic controllers and supervisory control and data acquisition systems were protected by their relative obscurity. Converging IT/OT networks and nation-state targeting, however, have changed the rules of the game.
To combat the escalating threat to manufacturing systems, many security operations centers (SOCs) are implementing agentic AI to preemptively identify and mitigate potential vulnerabilities. While this is a necessary development, our data shows that adoption of agentic security solutions is accelerating, but validation of their performance is not.
The Agentic Manufacturing SOC
To gauge how manufacturers and security operators are protecting vital OT systems, we surveyed 24 chief information security officers and senior security leaders working in OT environments between December 2025 and March 2026. We then augmented these survey findings with real-world performance data from within SimSpace testing environments.
Our data shows that agentic AI deployment in OT environments is on the rise, with 67% of respondents indicating they used agentic security solutions at least moderately and 8% identifying as power users with strong reliance on agentic defense. However, 25% of survey participants reported limited or no use of agentic solutions.
Very few manufacturers are implementing fully autonomous agentic security. Most see it as a means of enhancing the expertise of their security operators, with SOC augmentation, playbook automation, and tool optimization being the three primary areas in which agentic solutions are being deployed.
At SimSpace, we use a proprietary metric called defensive security readiness (DSR), which is a performance index evaluated as part of SimSpace team assessments that measures how effectively security teams detect and respond to adversarial activity during realistic intrusion simulations. DSR represents the proportion of attack activity that is correctly detected, investigated, and acted upon within a simulated environment.
DSR examines operational outcomes rather than traditional metrics, such as the volume of alerts generated, and instead evaluates how effectively human operators responded to simulated threats. Did the team or AI agent successfully identify the attack? Did they investigate it correctly? Did they take appropriate action? These criteria are continually assessed over time, giving security leaders insights into both current operational readiness and how performance improves – or declines – over time.
SimSpace data shows that, in tool onboarding scenarios, initial DSR can decrease by between 10-20 percentage points. This is typically caused by inexperience with new workflows and misconfiguration of specific tools. After between two and four iterative readiness exercises, however, DSR typically rebounds to exceed baseline performance by between 15-30%.
Data via SimSpace.
At face value, this is encouraging and indicates that agentic defense solutions are already influencing SOC performance. But without structured testing regimens and continual performance monitoring, the deployment of agentic AI initially introduces disruption and negatively impacts the very same efficiencies they’re designed to improve.
The Validation Gap in Agentic Defense
Our data shows that enthusiasm for and interest in agentic defenses in manufacturing OT environments is clearly on the rise. However, it also revealed the significant gap between perceptions of these systems’ effectiveness and the performance of actual teams.
We asked CISOs and security leaders to quantify their confidence in AI SOCs on a scale of one to five, with one representing minimal confidence and five representing maximum confidence. Approximately 75% of survey respondents rated their confidence in AI SOCs as either four or five out of five.
Data via SimSpace.
Security leaders clearly feel confident about the capabilities of agentic defenses, but this confidence may be misplaced.
SimSpace data reveals that teams with similarly high self-reported readiness maturity often display DSR variability as low as 0.30 across initial intrusion events. To complicate matters, this can often differ dramatically even within the same organization, with performance between individual events varying by as much as between 15-30 percentage points depending on scenario complexity, tool configuration, and team familiarity.
Data via SimSpace.
This tells us that while organizations may report high confidence, measured performance shows significant variability that is only uncovered through realistic simulations and ongoing testing.
Detection Speed is Increasing, Not Yet Fully Autonomous
One of the most attractive promises of agentic defense is the speed with which AI agents can accurately identify emergent threats. Our data shows that manufacturers using agentic defense solutions report significantly higher confidence in their ability to detect and respond to intrusion scenarios compared to the industry average.
Our survey revealed that 58% of survey participants reported median time to detection (MTTD) of between one and six hours, with OT operators being far more likely than the average organization (41%) to detect security incidents within six hours. One quarter (25%) of respondents reported a MTTD of less than one hour.
In terms of median time to response (MTTR), distribution is broader, with 42% of respondents reporting MTTRs of under one hour and between one and six hours respectively.
Data via SimSpace.
Our survey data shows that agentic defensive capabilities are closing the gap between detection and response in OT environments. Despite this, a significant proportion of OT organizations are not yet operating real-time response protocols, resulting in persistent measurement gaps. Data from the SimSpace platform explains why.
In early exercises, response precision and execution accuracy frequently lag detection performance; teams may detect threats, but fail to execute optimal response workflows consistently. However, with repeated testing in high-fidelity environments, response precision improves alongside detection, with typical improvement in response-related metrics ranging between 15-40% over successive events.
Data via SimSpace.
Detection is only half the battle. Agentic defenses may accelerate detection, but without repeated and rigorous testing, response execution remains inconsistent and dependent on the judgment of human operators. This represents one of the most urgently critical vulnerabilities facing OT SOCs today, and a major attack vector for PTAs.
Human Testing Failing to Match Pace of Agentic Security
Our data indicates that, for OT operators, the frequency of security testing is distributed fairly evenly across organizations that test continuously, semi-regularly, and on an ad hoc basis.
Data via SimSpace.
Many OT security leaders believe quarterly testing regimens are sufficient to ensure their operators remain poised to handle emergent intrusion events. However, this fails to consider the fact that agentic AI is operating continuously and that automated decision-making occurs daily. Most teams simply aren’t testing their capabilities frequently enough to match the pace of today’s agentic workflows.
SimSpace data shows how stark the differences in testing frequency and its impact on performance really are. Teams participating in frequent, full-scope events improve DSR by 0.03-0.05 per event on average, and typically achieve 0.70-0.80+ performance levels within four to six iterations. Conversely, teams using limited-scope or infrequent events improve more slowly and often plateau in the 0.50-0.65 range.
Data via SimSpace.
Continuous validation produces consistent, measurable gains, while infrequent episodic testing hinders performance and leaves significant gaps in AI-driven decision validation.
Closing the Agentic Confidence Gap in OT Environments
Continuous, on-demand training is more widely used in the manufacturing sector than industry averages. Despite this, our data reveals that tabletop exercises, certification programs, and tool-specific training remain the standard in many OT environments.
Conventional training still has its place across many aspects of modern manufacturing, but cybersecurity is not one of them. According to Forrester, traditional training modalities are insufficient[ii] when it comes to effectively identifying and responding to increasingly sophisticated AI cyberattacks. Forrester notes that organizations need training that “prepares practitioners to work alongside and evaluate AI-generated outputs for accuracy.” This means training programs must evolve to teach security professionals how to operate alongside AI agents, not just merely interact with them.
So how can today’s OT security leaders adapt to such a rapidly changing paradigm?
Shift from episodic to continuous testing
It is imperative that OT security leaders shift from organizational cultures of periodic testing to mindsets of constant validation to keep pace with the threats posed by agentic AI.
With agentic defenses operating continuously, ad-hoc training is simply no longer effective. Instead, CISOs in OT environments should implement continuous validation loops that encompass frequent, rigorous testing of detection, response, and AI-driven workflows. Organizations that test their security capabilities more frequently achieve measurably higher performance, while those that test less often plateau.
Focus on outcome-based performance metrics
One of the biggest organizational shifts that security leaders must spearhead is focusing on outcome-based performance data rather than legacy metrics that no longer apply in today’s SOC, such as detection alert volume.
Leaders must measure detection success, response accuracy, and decision quality across both human and AI workflows. Without outcome-based metrics, even the most forward-thinking organizations cannot accurately determine whether agentic AI is improving performance or introducing new risks across the attack surface.
Plan for the learning curve
As the tool implementation scenario data above demonstrates, AI adoption is far from frictionless. It takes time for human security practitioners to adapt to new systems of thinking and training, which can and does have a negative impact on performance in the short term.
Security leaders must take this into account when developing agentic training programs, and detection and response protocols. CISOs developing such programs must also ensure that continual validation becomes an institutional value at their organizations, as this leads to sustained, measurable performance gains over time.
Invest in AI Proving Grounds
One of the primary reasons why conventional training is no longer sufficient is that it cannot adapt to either the broader technical challenges facing OT security operators or the specific attack surface and threat vectors facing individual organizations. As agentic AI becomes integral to threat detection, investigation, and response, organizations must move beyond simply deploying automation and begin testing it under demanding, real-world conditions that mimic production environments as closely as possible.
This requires AI Proving Grounds, a realistic, high-fidelity replica of production environments in which organizations can simulate realistic adversary behavior, exercise AI-driven detection and response workflows safely, test automated playbooks thoroughly from start to finish, and evaluate how human analysts and AI agents perform together in actual intrusion scenarios.
Without AI Proving Grounds, organizations have little choice but to deploy agentic AI into production environments without fully understanding how it behaves under stress, where it fails, or how it interacts with human operators. This is especially risky within OT environments where maintaining production is crucial, and there are significant safety risks that could come from misconfigured machinery.
Preparing for Tomorrow’s Threats, Today
With manufacturing being the most-targeted industrial sector by PTAs for the past five years, the threats facing OT CISOs and security leaders have never been more urgent. Some organizations may be ready to respond to the next intrusion, but many are not.
Organizations that invest in AI Proving Grounds gain something far more valuable than sophisticated automations. They gain trust that their AI agents work, the confidence that their teams can rely on them, and the surety that their combined human-machine defenses will perform when it matters most.
[i] Why Manufacturing Cybersecurity is Becoming More Complex as Cyber Attacks Accelerate, April 2026
[ii] Announcing the Forrester Wave™: Cybersecurity Skills and Training Platforms, Q1 2026, April 2026
The content & opinions in this article are the author’s and do not necessarily represent the views of ManufacturingTomorrow
Featured Product
