Integrating AI into industrial systems creates new cybersecurity risks, requiring stronger access controls to protect legacy equipment and critical operations.
By Dr. Jaushin Lee, Founder and CEO, Zentera Systems
Integrating generative AI into industrial environments introduces severe cybersecurity vulnerabilities by connecting autonomous software agents to unpatched legacy operational technology (OT).
Because legacy industrial devices often rely on shared credentials and lack modern authentication, an AI agent with access to local systems can unintentionally or maliciously compromise critical physical operations in seconds. Industrial leaders must secure these environments by placing zero trust access controls directly around physical control assets rather than relying on legacy perimeter firewalls.
“Industrial control systems were built for operational longevity, not autonomous software interaction. When you introduce goal-oriented AI agents into environments running 20-year-old machinery, safety depends entirely on whether you can enforce non-negotiable operational boundaries around those physical assets.”
— Dr. Jaushin Lee, Zentera Systems Analysis
Industrial control systems and SCADA networks were engineered decades ago with a single primary mandate: operational uptime. Unlike standard enterprise IT environments where workstations are routinely updated and rebooted, OT systems run continuously for years or decades.
As a result, most deployed industrial infrastructure lacks modern authentication and authorization checks.
To keep operations smooth, plant engineers historically relied on shared administrative passwords, documenting them in operator logs or even on stickers on the appliance for convenience. When human operators managed these machines within air-gapped facilities, this practice posed a manageable risk.
Generative AI and autonomous agents completely dismantle that baseline. When developers or plant managers connect AI tools to industrial workflows — whether to analyze telemetry data, optimize output or write control scripts — those agents can search internal networks at machine speed. They can also get creative, deploying keyloggers or password crackers to achieve their goals.
If an agent discovers a shared password on an exposed endpoint, it gains full access to manipulate critical physical machinery.
Traditional software operates on rigid, predictable logic. Goal-oriented generative AI agents operate differently: You specify the destination, and the agent independently determines the execution path.
When an agent encounters an obstacle or code error while completing an assigned task, it explores available network corridors to resolve the issue. If strict network boundaries don’t constrain the agent, its autonomous search routines can inadvertently cross project lines, touch unmonitored programmable logic controllers (PLCs) or leak operational credentials.
According to the IBM “Cost of a Data Breach Report,” 13% of surveyed organizations have already experienced breaches directly involving AI models or applications, and 97% of those compromised organizations lacked proper AI access controls. In an industrial setting, the consequences of an AI-driven breach extend far beyond data loss. They threaten physical safety, supply chain continuity and critical public infrastructure.
Recent cyberattacks against regional water utilities highlight this precise vulnerability. State-sponsored threat actors exploited unpatched, internet-exposed industrial control units that still relied on default manufacturer passwords. If an autonomous AI agent or a malicious prompt injection leverages those same exposed credentials, the breach occurs before human operators even detect the anomaly.
To mitigate data exposure, executives may consider self-hosting open-source AI models as a conservative, secure choice. The assumption is that models run on local servers keep sensitive manufacturing blueprints and operational data safe behind corporate walls.
This assumption breaks down in practice, because self-hosting only controls where the model sits. It does not control where the AI agent can go. A self-hosted model running on a local server can still execute commands across your internal network, initiate unauthorized outbound socket connections or scrape adjacent databases.
Furthermore, self-hosting transfers all maintenance, vulnerability patching and threat monitoring onto internal IT teams. Unless you wrap environmental controls around the agent itself, self-hosting leaves your industrial supply chain exposed.
Industrial organizations cannot simply pause production to overhaul legacy networks or install complex agents on 20-year-old controllers. To safely adopt generative AI, manufacturers require non-disruptive, software-defined safeguards that drop directly over existing infrastructure.
Security teams must enforce zero trust principles at the asset layer rather than relying on network routing switches:
Legacy industrial control systems were designed for long-term operational availability rather than modern cyber defense. Many run on decade-old software that cannot be safely patched without risking costly operational downtime.
Further, these systems frequently utilize shared administrative credentials instead of individual user authentication, allowing an AI agent that uncovers a password to gain unrestricted control over physical machinery.
Manufacturers should deploy overlay security solutions that enforce zero trust access policies directly at the asset. By placing protective software boundaries around legacy control devices, organizations can strictly govern which users and application processes can communicate with physical machinery — without requiring complex network reengineering or physical hardware replacement.
Any industry relying on operational technology, SCADA networks or automated assembly lines faces elevated risk. This includes semiconductor manufacturing, pharmaceuticals, energy utilities, water infrastructure and heavy automotive production. In these sectors, an uncontained AI agent that crosses project boundaries can disrupt physical operations, compromise proprietary trade secrets or trigger severe regulatory penalties.
Integrating generative AI into industrial operations offers immense efficiency gains, but it cannot come at the expense of physical operational safety. Because legacy industrial machinery cannot be easily patched or re-architected, security must be enforced around the assets themselves.
By implementing overlay zero trust controls and strict process-level boundaries, manufacturers can more safely harness AI power while protecting critical production systems.
About the Author:
Dr. Jaushin Lee is the founder and CEO of Zentera Systems. He is a serial entrepreneur with many patents. He is also the visionary architect behind CoIP® Platform, Zentera’s award-winning Zero Trust security fabric. Jaushin has over 20 years of management and executive experience in networking and computer engineering through his experience with Cisco Systems, SGI and Imera Systems.
Read more from the author:
Why You Can’t Patch Your Way Out Of The Latest Water Utility Threats | Water Online, May 2026
Training Isn’t Enough: Cybersecurity in Manufacturing | Industry Today, October 2025
Forging the Next 250 Years: Powering the Next Era of American Manufacturing
Brooklyn Army Terminal (BAT) began as a World War II-era military supply base. Today it is New York City’s largest publicly owned industrial campus, with 59 waterfront acres, more than 4 million square feet and over 100 businesses. In this episode of Industry Insights, Bridget Chansakul, Vice President of Leasing & Portfolio Management at NYCEDC, explains what space is available and what’s planned for the 250,000+ square feet of new industrial space in Building A. She also covers how the ferry, parking and on-site childcare help tenants hire and keep workers, along with workforce partnerships, City incentives for manufacturers and where BAT is headed next.
Get In Touch
Google news and SEO compliant, Industry Today’s state-of-the-art digital media platform offers bespoke media campaigns that target key decision makers and buyers to achieve your marketing and promotional goals.
Contribute
Showcase your brand and promote your business to our highly targeted audience. We offer detailed Google Analytics with measurable ROI to assure success. Submit your content for review by our Editorial team who will contact you to discuss the project further.
About Us
Reach Your Targeted Audience and Grow Your Business. Learn more About Industry Today.
Contact Us
© 2026 Industry today. All Rights reserved.
