Checklist
Verify SCADA data and command paths before assuming an attack
Short answer
Most wrong or frozen SCADA values are tag, binding or driver faults, not attacks. Compare the display, the SCADA tag with its quality and timestamp, and the controller value: agreement clears the data path, and only unexplained writes or unauthorized reachability justify a cyber investigation.
On this page
A wrong or frozen value on a chemical-process screen is usually a tag, binding or driver fault, not an attack. Before anyone opens a cyber case, take one comparison at three levels: the value on the display, the SCADA tag behind it with its quality and timestamp, and the controller's live value read online. Where two adjacent levels disagree, the fault sits between them, and that is where you work. Only when all three agree and a change still has no authorized record, or a write appears with no attributable account, does it become a security investigation, structured the way IEC 62443 structures reachability: zones of assets with shared security requirements, and controlled conduits between them.
Compare three levels before anything else #
The comparison is taken at nearly the same moment at three levels: the value on the operator graphic, the SCADA tag that feeds it (value, quality code, last-update timestamp), and the controller value read from the engineering software in monitor mode. Read the controller without writing to it: forcing a value or toggling a rung on a running chemical process is a process-safety decision, and operations signs off on it before it happens.
Quality codes tell you which layer is complaining, so read them and the driver diagnostics before touching anything else. Ignition's 7.9 tag documentation lists a Stale quality with code 500 for a tag not evaluated within its expected time frame, and it points to the tag provider as the likely deeper problem. The same documentation is blunt about any quality that is not Good: the value should not be trusted, and the software overlays it so operators see it.
The trap is that quality can stay Good while the value is frozen; field-reported cases on Ignition show exactly that, which is why the timestamp belongs in the comparison. And when quality does go Bad, the most common cause is lost communication between the SCADA server and the device sourcing the tag — a path fault, not an intruder.
The rule that closes the data question: display, tag and controller agreeing within the tag's displayed resolution, with the timestamp current within one update period, clears the data path. A mismatch between two adjacent levels localizes the fault to the layer between them, and you fix that layer before any security work starts.
Five branches that localize the fault #
Work the symptom through the branch it matches. One wrong value with healthy neighbors is a binding problem; a group of values dying together is a shared driver or path; a command that moved by itself is the one branch that leads straight to the audit trails.
| What you see | What to compare | Where the fault is |
|---|---|---|
| One displayed value is wrong while neighboring values still update | Display binding, SCADA tag value, controller address | Binding, scaling or display logic — fix the graphic, not the network |
| Several tags stop updating together | Tag quality, driver diagnostics, connection logs | Shared driver, network path or controller connection |
| Controller value moves, the screen does not | Controller value, SCADA tag value, display timestamp | Acquisition (subscription or driver) first, then the binding |
| Screen and controller agree, the instrument disagrees | Instrument reading, controller input value, engineering-unit conversion | Instrumentation, input wiring or input configuration |
| A setpoint or command changed with no operator action | Operator audit records, controller change records, active sessions | Identify the writer; unexplained writes go to incident response |
A communication error is not evidence of compromise, and Good quality is not evidence that a command was authorized. Keep the two questions separate: is the data path intact, and who wrote.
Worked example: a frozen 78.4 °C display on a one-second reactor tag #
The case is typical for this fault class; the readings below are illustrative, while the quality-code reference is documented.
Inputs:
- Operator graphic: 78.4 °C, one decimal of resolution, no quality shown on the graphic.
- SCADA tag feeding it: 78.4 °C, quality Good, last-update timestamp 14:32:05.
- Controller value, read online in monitor mode: 81.0 °C, moving with the process.
- Wall clock at the check: 14:41:00.
- Tag subscription rate: one update per second.
- Documentation reference: Ignition codes a stale tag 500 — this tag read Good, so nothing flagged it.
Rule in words: subtract each adjacent level from the next, and compare the tag's timestamp age against its update rate. Any adjacent difference larger than the display's resolution means those two levels disagree.
Arithmetic:
- Display minus tag: 78.4 °C − 78.4 °C = 0.0 °C.
- Tag minus controller: 78.4 °C − 81.0 °C = −2.6 °C, which is 26 counts of the 0.1 °C display resolution.
- Timestamp age: 14:41:00 − 14:32:05 = 8 min 55 s = 535 s; at one update per second, 535 missed updates.
Result: the binding is correct and acquisition is dead between the controller and the SCADA tag. The tag froze at its last good value while quality stayed Good.
Decision: this is a data-acquisition fault, not a cyber case — the controller tracks the process and nothing wrote an unexplained change. Work the driver and subscription path, and add staleness detection so a tag quiet for more than a few update periods alarms. The comparison threshold that got you here: any adjacent difference above one display count (0.1 °C in this case) is a real disagreement.
Audit every path that can write #
With the data path cleared, the security question narrows to who can write and whether anyone did. Work the list in order:
- List every account that can write — interactive users, service accounts, privileged accounts, remote-access users — at the SCADA application, the operating system, the engineering software, the controller and the remote-access boundary.
- Compare each account's permissions with its operational task, and remove what the task does not require.
- Pull the operator audit trail and the controller change records around the time of the unexpected change; identify the writing account and the workstation it came from.
- Check where each credential is stored and which unattended services depend on it before changing or disabling anything.
- Use individually attributable accounts wherever the platform supports them; where a shared operator account remains, record the compensating controls that keep some accountability.
- Test one authorized action and one prohibited action with a representative role, and confirm both behave.
An unexpected command that maps to a known account is not closed — credentials are misused, so investigate the session and the workstation as well as the user. A write with no attributable record is the escalation trigger. Remote entries deserve the same rigor, which is why vetting remote access that leans on hardware keys is its own review, covered by the checklist to vet hardware key remote access for OT security.
Audit what can reach the controllers #
IEC 62443 gives the structure: a zone is a group of assets that share the same security requirements, and every connection between zones is a conduit that carries, and must control, the traffic crossing it. In a chemical plant the first cut is the control zone against everything else: business systems, remote access, external data exchange.
- Read the firewall rules, the routing configuration and the remote-access design together. Separation on a drawing is worthless if routing still lets an unrelated system initiate a connection to a controller.
- Group assets into zones by trust and operational role, and permit between zones only the communications a documented function needs.
- Confirm from configuration readback, not from the diagram, which systems can open a session to each controller.
- A firewall limits traffic; it does not make an authorized-but-compromised workstation harmless. Access control and endpoint protection still apply inside the permitted path.
The same reachability review sits behind segregating machine drive traffic from the plant network and connecting a PLC to the plant network without exposing drives; done well, it is why layered OT security controls across a network path hold up under a real incident.
When it becomes a cyber case #
Commit on what the checks produced:
- Any mismatch between adjacent levels: a data-integrity fault. Fix the layer between them; no security escalation on that evidence alone.
- All three levels agree, and a value or command changed with no authorized record in any audit trail: treat it as suspected compromise. Preserve the audit records and controller change logs, restrict the write path you identified, and hand it to the incident-response process.
- Values moving with no process cause while all three levels and the instrument agree: verify the controller logic and its change history before anything else.
Knowing the attack types that actually hit industrial control networks shapes the briefing you give management, because containment differs for each.
Questions asked
Does a Good quality code mean the displayed value is current?
No. Quality reports whether the tag evaluated on time, not whether the value is fresh or authorized. Field cases on Ignition show tags frozen at their last value while quality still read Good, so alarm on the last-update timestamp as well: key the alarm to the tag's update rate so a one-second tag alarms within seconds of going quiet, not hours.
Is a stale-tag quality code of 500 the same across SCADA packages?
No. Code 500 with that meaning is documented for Ignition 7.9 tags. Quality codes and names differ between platforms, so read your platform's own documentation, record the codes your drivers actually raise during normal faults, and put those codes in the diagnostic table.
Should we scan the control network to verify what can reach the controllers?
Not against a running system without planning. Active discovery can place unexpected load on legacy control devices. Configuration readback of firewalls, routing and remote-access design answers the reachability question; if a scan is still required, run it in a window operations agrees can tolerate a disruption.
If the audit trail names an operator account, is the investigation finished?
No. Credentials get misused, so the record is the start of the investigation. Check which workstation and session issued the write, whether that operator was on shift at the time, and which other write paths bypass the audit trail entirely. A missing record is not proof that nothing happened.
Where does IEC 62443 say to start in a chemical plant?
With zones and conduits. Group assets that share security requirements, controllers and their SCADA servers in one zone, business systems outside it, and treat every connection between zones as a conduit that is documented and restricted. The control-to-business boundary and the remote-access path are the first two conduits to tighten.