Critical Infrastructure sites are rightly among the most heavily guarded places in industry, with many independent layers of clearance between the front gate and an HMI. That depth of defence is the reason this piece exists, not something it argues against.
Let's imagine...
Picture a plant built to defence-in-depth standards: perimeter fencing with a manned gatehouse, vetted security clearance for everyone on site, badge-controlled turnstiles, escort-only access to the process areas, a locked control room with a two-person rule, network segmentation, hardened HMIs and USB port blocking, and a policy for every one of these steps, signed, audited and compliant. On paper, nobody is getting through.
Now look at that list again. Behind every gate there's a person: a guard who decides whether a story checks out, an escort who decides whether to step away “just for a minute”, an engineer who decides whether to plug in a vendor's laptop, an operator who decides whether to accept an “urgent OEM patch” over the phone.
A process holds until somebody decides to override it, usually for a reason that seems sensible at the time: a deadline, being polite, a familiar face, or what looks like an emergency. An attacker doesn't have to defeat ten gates. They only need to persuade the people who look after them.
The comfortable assumption of "compliant"
During project implementation, the controls get proven on paper twice over. FAT shows the system behaves correctly in a clean, controlled factory. SAT shows it behaves correctly again once it reaches the field.
Neither one checks whether the system, or the people around it, can be persuaded to fail. A signed compliance checklist says a rule exists; it says nothing about whether that rule survives contact with a plausible excuse at 3 a.m.
Rules can't be expected to work until they've been tested at random, under the same pressure and ambiguity an attacker would actually use against them.
One pentest, checked at two points
The recommendation is a single end-to-end pentest engagement, run at two stages of the project. Post-FAT: a full technical pentest, testing the system in depth in the lab, where it's cheap and quick to do. Remediation then closes every finding before the system goes live.
Post-SAT: a targeted re-test plus a covert human assessment, checking the system as installed, connected and run by real people, not the clean lab build. The result is a sealed handover — the client takes over a system whose security, including the human layer, has been proven with evidence rather than assumed from a checklist.
"Isn't testing twice a waste? Just do it after SAT."
Only exists post-SAT
- Site integration — client IT/OT network, historians, DMZ, remote access, AD
- Commissioning drift — debug ports, test accounts, "allow any (temp)" firewall rules
- Configuration changes made on site after FAT
- Physical layer — unlocked cabinets, exposed switch ports, vendor laptops, 4G routers
- Human layer — real guards, real operators, real contractors under real pressure
- Regressions — post-FAT fixes undone during installation
What post-FAT already proved
- Core logic and configuration behave correctly in a controlled lab
- Most defects are cheap to find and fix before site mobilisation
- Vendor ownership of findings is unambiguous under warranty
- Test cases and asset knowledge carry forward to the site visit
Cost of testing once vs. testing twice
| Approach | Pentest spend | Cost of fixing findings | Schedule risk | Total exposure | ||
|---|---|---|---|---|---|---|
| Post-SAT only | 1.0x | High — remobilisation | outages | MoC | High — can block handover or trigger LDs | Highest |
| Post-FAT + post-SAT | ~1.3-1.5x | Low — most fixes made in the lab | Low — findings come off the critical path | Lowest |
The cost goes up a little and the risk goes down a lot
If the post-SAT test finds nothing new, the earlier spend wasn't wasted — it's evidence that remediation held, and that evidence is what makes the handover sealed. The same finding can cost ten to a hundred times more to fix after SAT than it would have in the lab.
The increase stays modest for four reasons: post-FAT does the heavy lifting where testing is safe, aggressive and cheap; post-SAT is shorter and targeted — site drift, a check that fixes held, and the human assessment; it's one package with one team, so mobilisation and test cases carry forward; and ownership stays clear, since post-FAT findings belong to the vendor and get fixed under warranty rather than argued over at handover. A single critical finding discovered after SAT can cost more than the whole post-FAT test.
Testing at both stages raises the pentest budget by a fraction and cuts the cost of fixing findings by a multiple. It's planned spend in place of an emergency later.
The human extension
Pentesting isn't only about devices — people need to be part of the loop, because processes protect them less than we think when people can always override the process. A covert social engineering assessment should be a standard extension of every OT pentest package, not an optional extra.
Impersonation
Posing as a contractor, OEM engineer, auditor or courier to test whether a story is checked before a door opens.
Vishing
"Vendor support here, we need remote access for an urgent fix" — testing whether urgency alone is enough to bypass verification.
Targeted phishing
Aimed at engineering and operations staff, not just the general workforce.
Tailgating and escort-policy tests
Checking whether an escort ever steps away "just for a minute".
Planted media
Left in a plausible location, to see whether anyone reports it instead of plugging it in.
The test must be random and unannounced. If people know a test is coming, you measure how they behave when watched — that isn't how they'll behave on a normal Tuesday at 3 a.m.
Nobody is named, shamed or disciplined. Results are anonymised and reported as process and training gaps, not individual failures. Punish people and they'll hide mistakes; coach them and they'll report. The aim isn't to catch people out — it's to make them more resilient.
Why this matters
Critical Infrastructure isn't short of technology, policy or clearance levels. What usually breaks is the moment when a well-meaning person decides that this one time, the rule can wait. A pentest that covers only devices is half a pentest, and a pentest run at only one stage leaves gaps — test the system in the lab, check it again on site, and include the people who run it.
When did your organisation last test the people behind the gates, rather than just the gates themselves?
Would you treat a two-stage pentest as a waste of money, or as insurance?