Securing operational technology: identity governance in the age of AI agents and industrial ransomware

AI Agents, Microsegmentation, and the Expanding OT Threat Landscape

The line separating digital workflows from physical infrastructure is disappearing. Enterprises are deploying autonomous AI agents while simultaneously connecting sensitive operational technology environments to broader corporate networks. As attackers exploit both machine identities and industrial dependencies, security teams are being forced to unify IT, OT, and AI governance under a single operational model.

Perimeter defenses alone are no longer sufficient. Enterprise resilience increasingly depends on continuous identity verification, hardware-rooted trust, and deep visibility across interconnected systems.

Advancements in Industrial Microsegmentation Security

Next-generation segmentation combines artificial intelligence with hardware-backed trust to contain breaches across converged IT and OT environments.

The convergence of Information Technology and Operational Technology is accelerating a major shift in industrial security strategy. With ransomware disrupting manufacturers like Jaguar Land Rover and NIS2 regulations increasing compliance pressure across the European Union, organizations are rapidly reevaluating how they isolate and protect operational assets.

Industry forecasts project the microsegmentation market will grow at a 23.70 percent compound annual rate through 2031, reflecting broader enterprise adoption of zero-trust architectures in hybrid environments. Zero Networks reported an 80 percent year-over-year increase in OT customers across manufacturing, energy, utilities, and transportation. The growth signals strong demand for containment technologies that protect production uptime without operational disruption.

Traditional segmentation relied heavily on static firewall rules and manually maintained network boundaries. Those approaches struggle in modern industrial environments where devices, workloads, and remote access requirements change continuously.

Modern microsegmentation platforms instead analyze communication patterns between engineering workstations, programmable logic controllers, monitoring systems, and administrative infrastructure. Machine learning models establish behavioral baselines and dynamically restrict unexpected east-west traffic, helping reduce lateral movement without requiring constant firewall reconfiguration.

Security teams are increasingly pairing software-defined segmentation with hardware-rooted trust. Architectures aligned with the Purdue model now incorporate Trusted Platform Modules and Trusted Execution Environments to validate endpoint integrity before devices communicate across broader operational networks. This combination of identity verification, segmentation, and hardware attestation helps isolate legacy industrial systems while improving visibility across converged environments.

The Operator’s Lens

• Prioritize segmentation platforms that automate policy generation instead of expanding manual firewall administration.
• Require TPM-backed endpoint validation before operational assets connect to broader enterprise networks.
• Evaluate vendors based on proven hybrid IT and OT deployment experience rather than generalized zero-trust marketing claims.

Bridging the gap between physical infrastructure security and software governance requires the same disciplined identity controls now reshaping enterprise AI deployments.

Orchestrating the Autonomous Enterprise

Deploying AI agents safely requires identity governance frameworks that bind autonomous actions to accountable human oversight.

Google recently expanded CodeMender from a standalone remediation tool into a broader agentic development ecosystem within its Gemini Enterprise platform. The move reflects a broader enterprise reality: organizations increasingly expect autonomous tooling to operate inside governed environments that include centralized identity management, gateway controls, telemetry, and approval workflows.

AI-native remediation pipelines are increasingly being used to identify and patch vulnerabilities at machine speed. At the same time, enterprises are confronting the rapid emergence of shadow agents.

Developers are increasingly deploying AI coding assistants and Model Context Protocol servers without formal security review. These agents can retain persistent API access, read sensitive repositories, invoke external services, and execute automated remediation tasks outside centralized monitoring pipelines.

In many cases, security teams only discover unauthorized agents after they begin interacting with production systems or transmitting proprietary data externally.

Securing these environments requires more than traditional endpoint management. Organizations are deploying behavioral telemetry, endpoint-level inventories, and machine-speed policy enforcement to track data lineage, restrict unauthorized access, and reduce prompt injection risks that could manipulate autonomous workflows.

Security architects are also treating AI agents as non-human identities subject to the same governance expectations applied to privileged users and service accounts. Identity-first frameworks increasingly rely on fine-grained authorization models, short-lived credentials, and secure secrets management rather than static API keys.

The emerging “agent rule of two” governance model links each autonomous credential to a verifiable human owner, helping establish accountability for machine-driven actions across enterprise systems.

Infrastructure flexibility remains equally important. Regulated industries increasingly require private cloud or air-gapped deployment options to protect sensitive data during AI synchronization and orchestration workflows.

Human-in-the-loop governance is also becoming standard practice. Cisco XDR, for example, uses agentic AI to automate incident verification and summarize attack chains while reserving final response authority for human analysts. This model allows organizations to increase operational speed without fully surrendering decision-making control.

The Operator’s Lens

• Shift software security spending toward integrated platforms that enforce centralized identity governance across autonomous tooling.
• Replace long-lived agent credentials with just-in-time access tokens and tightly scoped permissions.
• Train identity and access management teams to govern AI coding assistants as non-human privileged entities.

As enterprises expand autonomous workflows internally, threat actors are simultaneously exploiting these same infrastructure dependencies externally.

Evolving Operational Technology Cyber Threats

Advanced threat groups are moving beyond traditional IT perimeters to establish persistence inside operational environments and critical supply chains.

The operational technology threat landscape continues to evolve rapidly. According to the 2026 Dragos, Inc. OT/ICS Cybersecurity Report, advanced adversaries are increasingly operating as Stage 2 actors inside industrial environments, where they actively map processes, probe network relationships, and analyze industrial control system communications.

Dragos also identified several specialized threat groups operating across coordinated intrusion chains. Groups such as SYLVANITE focus on gaining initial access through exposed internet-facing systems before handing operations to actors focused on data exfiltration or destructive payload deployment.

Ransomware operators are refining their industrial targeting strategies as well. In 2025, 119 ransomware groups reportedly targeted industrial organizations, representing a sharp increase from the previous year.

Rather than directly manipulating industrial equipment, many affiliates now target the virtualization infrastructure supporting OT operations. Historians, SCADA visualization systems, engineering management platforms, and operational monitoring workloads increasingly depend on centralized hypervisors. Encrypting those systems can sever operator visibility across entire facilities without directly touching physical control equipment.

Engineering workstations also remain a major blind spot. Many organizations continue classifying operational engineering software as standard IT infrastructure, obscuring the true operational impact of ransomware incidents and expanding the attack surface available to adversaries.

Despite rising awareness, many organizations still struggle with foundational readiness. Recent Dragos assessments found that 88 percent of tabletop exercises revealed degraded detection capabilities, while 81 percent of architecture reviews exposed inadequate IT and OT segmentation. These visibility gaps contribute to an average adversary dwell time of 42 days inside operational networks.

Building resilience requires organizations to test entire dependency chains rather than isolated security boundaries. Accurate asset inventories, continuous network monitoring, segmentation validation, and OT-specific incident response planning are becoming foundational operational requirements.

The Operator’s Lens

• Conduct tabletop exercises that simulate virtualization outages and supply chain compromises across operational environments.
• Deploy continuous monitoring deeper within OT networks to detect reconnaissance and process mapping activity earlier.
• Partner with incident response firms that possess demonstrated industrial control system expertise instead of relying exclusively on traditional IT security providers.

The convergence of autonomous AI systems and vulnerable operational infrastructure is forcing a broader transformation in enterprise security strategy. Organizations can no longer treat IT, OT, and AI governance as separate disciplines.

Future resilience will depend on unifying identity-centric security, infrastructure visibility, and operational segmentation into a governance model capable of protecting both machine-driven workflows and physical industrial systems.