Unauthorized network access
A new device talking to a PLC, an IT workstation reaching an OT controller, an unexpected protocol, port or device pair, or traffic at the wrong rate or time.
We find activity that is technically valid and still wrong – early enough to investigate before it becomes an operational, safety or business incident.
Understand what exists, what is allowed to communicate, what each device is expected to do, and what counts as abnormal.
Conventional intrusion detection looks for traffic that is known to be bad. In a plant, the damaging command is often a perfectly valid one – sent from the wrong place, at the wrong time, to the wrong controller, with a physical consequence nobody connected to it. So we model the environment first, and judge every event against what that environment is supposed to be doing.
A new device talking to a PLC, an IT workstation reaching an OT controller, an unexpected protocol, port or device pair, or traffic at the wrong rate or time.
A write to a controller that never receives writes, a valid command from an unexpected source, or a sequence of individually valid commands that should not occur together.
Parameters pushed outside the operating envelope, sensor values that disagree with the rest of the process, or a cyber event followed by an abnormal physical response.
Impossible or inconsistent readings, a changed sensor identity or configuration, or a device drifting from its own historical baseline.
Unknown network devices, unexpected controllers, or a model, firmware or configuration that does not match the expected device profile.
Previously isolated devices starting to talk, new communication paths appearing, or one endpoint scanning or reaching across several controllers.
This is where we differ. We correlate the cyber event with what the device did, the state of the process, and the physical consequence. An attack that looks normal in network traffic alone stops looking normal once the process is in the picture.
A source that has never reached this zone connects to it.
It issues a command the controller does accept.
A setpoint moves – still within what the protocol allows.
The process responds, as it was told to.
The physical system stops behaving the way its history says it should.
Read together, five unremarkable events become one incident.
Risk is contextual. A new device, plus access to a controller, plus an unusual command, plus an abnormal temperature change is one high-risk incident – not four alerts for an operator to stitch together.
Three sources of intelligence feed one detection engine, which reports to your operators. Normal behaviour is learned at four levels – network, device, identity and process – so a deviation is always a deviation from something specific.
TAP / SPAN, flows, firewall and infrastructure logs, and protocol metadata for Modbus TCP, OPC UA, MQTT, EtherNet/IP and PROFINET. Protocol-agnostic, so new parsers can be added.
What each device is, where it sits, what it runs, what it talks to, what it is expected to do, and which physical process it affects – kept as a living asset and relationship graph.
Pressure, temperature, current, flow, vibration, RPM, valve and motor state – whatever the process exposes, through the historian or SCADA you already have.
Deterministic rules for known violations, behavioural baselines for drift, and contextual risk scoring that combines weak signals instead of alerting on each one.
Each detection carries the assets involved, source and destination, the observed behaviour next to the expected behaviour, the policy it touches, the related process signals, a risk and a confidence level, and suggested next steps. Related events are grouped into incidents.
Operators investigate through an environment map of assets, zones and communication paths, an incident timeline from first access to physical response, and a full history for every asset. For every alert the system answers two questions: why was this considered abnormal, and what changed compared with the baseline.
AI sits on top of this rather than in place of it – natural-language investigation, incident summaries, correlation of weak signals and help writing detection rules, all inside strict access and audit boundaries.
Monitoring from a TAP or SPAN port. The system never sits in the control path.
Watching and acting are separate functions with separate permissions.
Local collectors and a private platform. Cloud is optional and yours to decide.
Least privilege, role-based access, strong identity, encryption and immutable security logs.
AI assists investigation. It has no uncontrolled access to critical OT systems.
Blocking or automated control changes only after rigorous safety validation and explicit approval.
It augments your security environment rather than replacing it: connectors for SIEM and SOC, IAM, firewalls, asset management, SCADA / HMI, historians, existing OT security platforms, ticketing and IoT platforms.
Not with every possible attack. A first deployment covers a small number of high-value scenarios, validated against concrete attack scenarios you give us.
Passive network discovery and an automatic asset inventory.
The asset relationship graph and a baseline of normal communication.
Unauthorized communication, unexpected devices, command anomalies, and configuration or firmware changes.
Selected process and sensor data, joined to cyber events into risk-ranked incidents.
A dashboard for operators and an integration into your SIEM or SOC.
Success is not the number of attacks detected. It is whether meaningful deviations surface early enough for you to investigate and respond.
We choose technology and architecture after we understand your environment, not before. The first conversation covers: