Security Operations for Critical Infrastructure
How OT-aware security operations differs from a conventional SOC: what to watch, how to build detection use cases that matter, how to triage without drowning in noise, and where telemetry actually comes from.
What a SOC Actually Watches in OT Environments
A security operations program built for IT and pointed unchanged at an OT environment usually fails quietly: it generates alerts nobody trusts, misses the events that actually matter, and burns analyst attention on noise. Watching OT correctly starts with accepting that the environment behaves differently and needs to be monitored differently. OT networks are far more predictable than IT networks in normal operation -- the same PLC talks to the same HMI on the same schedule, day after day, because industrial processes are repetitive by nature. This predictability is an advantage: it means baseline deviation (a new device appearing, a command sent outside the normal pattern, communication to an unexpected destination) is a much stronger and cleaner signal in OT than the equivalent would be on a chaotic corporate IT network full of legitimately varied user behavior. A mature OT-aware SOC watches for categories that a conventional IT SOC often does not prioritize: new or unauthorized devices appearing on the OT network, engineering workstation activity outside approved change windows, command sequences to controllers that fall outside normal operational patterns, communication attempts across segmentation boundaries that should not exist, and authentication activity on remote access paths used by vendors and third parties. Just as important as what to watch is what not to over-alert on: a SOC that treats every protocol anomaly with the same urgency as a confirmed unauthorized command will exhaust its analysts and lose credibility with operations teams, who will start dismissing alerts by default -- which is precisely the failure mode a security program cannot afford in a critical facility.
Building Detection Use Cases Around Real Threats
A detection use case is only valuable if it is built around a specific, credible threat scenario -- not a generic 'detect malware' aspiration that nobody can actually validate or tune. The most effective OT detection programs work backward from realistic attack scenarios, often grounded in frameworks like MITRE ATT&CK for ICS, rather than forward from whatever data happens to be easiest to collect. A well-formed use case names the threat scenario explicitly -- for example, 'unauthorized engineering workstation used to push a configuration change to a PLC outside an approved maintenance window' -- and then defines exactly what data source would reveal it (engineering workstation logs, PLC command logs, change management system records), what the detection logic actually looks for, what a true positive looks like versus a false positive, and what the response action should be if it fires. Building use cases this way forces a useful discipline: if you cannot name the data source that would reveal the scenario, you do not actually have detection coverage for it, no matter what your security architecture diagram claims. This is one of the most common gaps auditors find -- a control that looks complete on paper but has no corresponding, tested detection logic behind it. Prioritization matters as much as construction: a small number of high-confidence use cases tied to your organization's most consequential threat scenarios (informed by your risk register, not a generic industry list) will deliver more real security value than a large number of shallow, low-confidence rules that mostly generate noise.
Alert Triage and Escalation Without Alert Fatigue
Alert fatigue is not a minor inconvenience for a SOC -- it is a direct security failure mode, because an analyst who has learned to dismiss alerts by default will eventually dismiss the one that mattered. In OT environments, where a missed critical alert can have safety or availability consequences, controlling alert volume and quality is a security control in its own right, not just an operational efficiency concern. Effective triage starts with tiering: not every alert deserves the same response speed. A confirmed unauthorized command to a safety-critical controller demands immediate escalation; a low-confidence protocol anomaly on a non-critical monitoring segment can be queued for review during business hours. Programs that treat every alert with uniform urgency train analysts to stop distinguishing between them, which defeats the purpose of tiering entirely. Context enrichment dramatically improves triage speed and accuracy: an alert that arrives with the asset's criticality, its normal behavior baseline, its current maintenance status, and its owner already attached can be assessed in seconds. The same alert with no context requires the analyst to manually chase down all of that information first -- often taking longer than the investigation itself. Escalation paths need to be explicit and rehearsed before an incident happens, not designed during one. Who gets called for a confirmed OT security event, at what severity, through what channel, and what operational authority does that person have to request a process change or isolation action -- these questions should have clear, tested answers, because discovering the answer live during an active incident costs time that a safety-relevant scenario may not allow.
Telemetry Sources: Logs, Network, Endpoint, and Historian Data
Detection is only as good as the telemetry feeding it, and OT environments have data sources that a conventional IT-focused SOC often overlooks entirely -- while also lacking some sources IT security teams take for granted, like widely deployed endpoint agents on legacy controllers. Network telemetry is usually the richest and least intrusive source in OT, because passive network monitoring (via a network tap or span port) can observe traffic and protocol behavior without touching the controllers themselves -- critical, since many legacy devices cannot safely run additional software or tolerate active scanning. Deep packet inspection tuned for industrial protocols (Modbus, DNP3, OPC, and others) can reveal specific commands, not just traffic volume, which is what enables detection use cases like 'unauthorized write command to a specific register.' Endpoint telemetry is available where it makes sense -- engineering workstations, HMIs, SCADA servers, and jump hosts typically run modern operating systems that can support logging and, where operationally appropriate, endpoint detection agents -- but should never be assumed available on the controllers and field devices themselves. Historian data -- the time-series process data every industrial facility already collects for operational reasons -- is an underused security asset. Because historians already record process values over time, they can reveal physical-layer consequences of a cyber event (a pressure or flow reading that deviates in a way that correlates with a suspicious network event) that no purely cyber-focused telemetry source would catch on its own. Correlating cyber telemetry with process telemetry is one of the more advanced but genuinely valuable practices a mature OT SecureOPS program can build toward.