Manufacturers race to adopt AI while security teams struggle to manage the risks
Manufacturers are embedding AI agents into production scheduling, quality inspection, logistics, and predictive maintenance at scale-but most security teams lack visibility into what these systems are doing and what damage they could cause if compromised.
Darktrace's latest research found that 78% of manufacturing security professionals worry most about employee use of AI agents. That concern outpaces worries about generative AI tools like ChatGPT (76%) and reflects a fundamental problem: AI agents operate with broad permissions and minimal human oversight, making them attractive targets for attackers and potential sources of unintended harm.
Unlike traditional AI that performs predefined tasks, agents make decisions autonomously and adapt to new situations. They can access enterprise systems, interact with external platforms, and execute complex operations without human approval at each step. "They look like employees operationally, but lack judgment, ethics, or fear of consequences like humans do," Darktrace said in its analysis. "This means they can be easily manipulated by cybercriminals."
Attackers are already using AI to move faster
The threat isn't hypothetical. Seventy-six percent of manufacturing security professionals surveyed report already being hit by AI-powered attacks. Ninety percent expect AI to increase the success of social engineering campaigns.
Attackers now use AI to automate reconnaissance, refine targeting, and adapt attacks in real time. What once required manual coordination can now run continuously at scale. Hackers identify vulnerabilities faster, craft more convincing phishing emails, and move across interconnected IT and operational technology networks more efficiently.
Manufacturing teams are most concerned about three attack types:
- Adaptive malware that evolves in real time (49% worried)
- Automated vulnerability scanning and exploit chaining (48%)
- Hyper-personalized phishing campaigns (46%)
Most manufacturers aren't ready
Fifty-one percent of manufacturers say they're not adequately prepared for AI-driven threats. Only 37% have formal policies governing AI deployment.
The gap between adoption and readiness creates operational risk. Sixty percent of security professionals worry about sensitive data exposure through AI systems. Fifty-nine percent fear accidental policy and regulatory violations.
Today, most organizations have minimal visibility into how AI agents behave in their environments. Ninety-one percent of manufacturing security professionals said they need to understand how AI makes decisions before trusting it-especially critical in operational settings where system failures can affect safety, environmental compliance, and finances.
Three priorities for securing AI in manufacturing
Visibility: Organizations cannot secure systems they don't understand. Teams need to know where AI is deployed, what data it can access, and how it behaves across both IT and operational technology networks.
Context: In environments shaped by AI, normal behavior constantly shifts. Security teams need to detect threats by understanding patterns of life across the organization and spotting subtle deviations in real time-a departure from traditional security approaches.
Guardrails: As AI systems take on greater responsibility, organizations need clear boundaries around what they can do and when they can act independently. These controls must be built into systems themselves, not applied afterward.
The core challenge: security must operate at the same speed, scale, and complexity as AI itself. Slowing innovation isn't the answer. Building visibility and controls into AI systems from the start is.
Operations leaders deploying AI in manufacturing should review what visibility their security teams actually have into these systems and whether formal policies exist before expanding AI use further. The cost of discovering a compromised AI agent during production is far higher than investing in controls beforehand.
Your membership also unlocks:
