Shutdown at 3:40 a.m.: When OT and SCADA security only becomes visible at restart
No ransom note, no SIEM alert, no flashing red light. Just a control room where the process displays freeze, and a shift crew that can’t tell whether the plant is still running safely. This is the moment that shows whether OT and SCADA security holds up in practice or only exists on paper, and according to the latest SANS survey, it now happens to one in three industrial organizations.
The sequence is almost always the same. The night shift reports that the HMI screens have hung. IT checks whether it’s the network. The automation engineer checks whether it’s the controllers. Forty minutes later it’s clear that the virtual machines running the HMI, the historian and the control system have been encrypted. The PLCs themselves are still working. It’s just that nobody can see them anymore.
Then comes the phase hardly anyone has a playbook for. Do you bring the plant down in a controlled shutdown because nobody knows how far the attacker got, or keep it running in manual mode, which someone last practiced in 2019? The person who has to make that call usually isn’t the CISO. It’s a shift supervisor, a few minutes before four in the morning.
The figures from the SANS 2026 survey show that this scene has long since become the statistical norm after an incident in an operational technology environment.
Industry has partly solved its detection problem and hasn’t even touched its recovery problem. Nearly three out of four reported incidents ended in a plant shutdown in 2026.
What the 2026 SANS survey actually measures, and what it doesn’t
The study has surveyed security leaders from industrial environments every year since 2017. Nearly 700 professionals from virtually every critical infrastructure sector responded to the current edition. This isn’t sensor telemetry; it’s self-reported data. That makes the jump from 22 % to 34 % in a single year all the more striking: it’s the largest year-over-year increase in the survey’s history.
At the same time, detection speed has slowed down again for the first time, and in the very year when more is happening. The industry has spent years investing in monitoring, yet it’s now seeing attacks coming more slowly.
| Metric | Value | What it means |
|---|---|---|
| Incident rate 2026 | 34 % | High Plan for an incident; it’s not the exception |
| Share with downtime | 72 % | High Protection works, resilience doesn’t |
| Detection within 24 h (2025) | almost half | Solid Detection isn’t the bottleneck |
| Remediation over 30 days (2025) | 19 % | High Recovery is the bottleneck |
| Budget as a barrier | 43 % | Top answer ahead of staff and technology |
In Germany, many people tend to overlook one finding from the previous year’s survey. Regulated sites had roughly as many incidents as unregulated ones, but almost 50 % less financial and safety impact. Regulation doesn’t prevent the attack. What it apparently does is shorten the time it takes to regain control.
The ICS Cyber Kill Chain and why stage two matters
The model splits industrial attacks into two stages. In stage one, the attacker gains access to IT and to the network adjacent to OT. Only in stage two do they learn the process, develop an effect and trigger it. Defenders need visibility into both stages to be able to say where an attack stands when it counts. According to SANS, only 13 % of respondents manage this end to end. If you can’t see into stage two, shutting down is the safe default.
Why almost three quarters of incidents bring production to a halt
The path to a shutdown almost never runs through a direct attack on a controller. It runs through systems that are technically IT but organizationally OT: virtualization hosts, remote access gateways, historian servers, the integrator’s engineering workstations. In many plants, nobody really owns this layer.
That fifth step is the expensive one. Technically, it doesn’t necessarily follow from the attack; it’s a decision under uncertainty that someone has to make. And the less you know about the process level, the more conservative that decision will be. If you never have the links between the office network, remote maintenance and the cell level tested in practice, you’ll discover them for the first time during a real incident. An OT pentest answers the questions in advance that nobody has time to research at four in the morning.
Clorox showed in 2023 that even purely upstream systems can stop a supply chain. After its automated ordering systems went down, the company reported a 20 % drop in revenue and total costs of 356 million US dollars.
Three assumptions that make every recovery more expensive
- IT SIEMs usually don’t understand Modbus, S7 or Profinet
- The first sign is often the shutdown itself
- Only 13 % can see the full chain
- ~50 % detected within 24 h
- 60 % contained within 48 h
- Yet 19 % remain open for over a month
- Manual operation is considered the most glaring resilience gap
- An unpracticed switchover becomes a bottleneck in its own right
- Only one in ten could manage long-term without their main vendor
Vendor dependency deserves a closer look of its own. If patches, spare parts, license servers and remote support all hinge on a single automation vendor, your recovery time is effectively their response time. Lawmakers are moving in the same direction. Since 2026, operators of critical facilities have had to report the types of critical components they use to the BBK and BSI when registering under Section 33(2) BSIG; prohibitions are issued after the fact under Section 41 BSIG.
Prevention and operational resilience are two separate budgets. If you only fund the first, you’ll pay for the second later in hours of downtime.
Recovery planning for OT: five measures in order of priority
The following points are ranked by their impact on downtime, not by how much effort they take to implement. In most plants, points one and two can be tackled within a quarter.
- Practice recovery, don’t just document it: An ICS playbook covering a full production restart, including the order in which plant sections come back online, sign-offs from the process owners and acceptance criteria. Run it as a live exercise at least once a year, not as a tabletop.
- Test emergency operation for real: Manual operation with a real shift crew, real handwheels and real time pressure. If you only describe it, what you have is an assumption, not a plan. SANS counts this among the most glaring resilience gaps.
- Visibility down to the process level: Asset inventory, network capture and protocol analysis all the way down to the cell level. The goal is to be able to prove within hours during an incident that the control layer is clean.
- Harden and segment OT-adjacent IT: Virtualization hosts, jump hosts, remote maintenance gateways and the historian get their own zones, their own accounts and separate backups. This layer was the entry point in the case analyzed by Dragos.
- Get the reporting chain down to 24 hours: Operators of critical facilities must report significant incidents to the BSI within 24 hours, followed by staged follow-up and final reports. This escalation belongs in the same exercise as the restart, because it falls on the same people. A KRITIS audit shows what proof of effective measures looks like in practice.
Don’t set your target metric to mean time to detect; set it to mean time to production. Anything else measures the wrong end.
Questions about OT incidents, downtime and recovery
How many industrial companies had an OT security incident in 2026?
According to the SANS 2026 State of ICS/OT Security Survey, 34 % of the nearly 700 industrial security professionals surveyed reported a cyber incident in their control system or operational technology environments. The previous year’s figure was 22 %, making this the largest year-over-year increase since the survey began in 2017.
Why do so many OT incidents lead to production outages?
72 % of reported incidents in 2026 resulted in plant or system downtime. The main causes are a lack of visibility across IT and OT, which only 13 % of organizations achieve end to end, and unpracticed emergency operating modes. Without a reliable picture of how far an attack has spread, operators shut down when in doubt.
How long does recovery take after an ICS incident?
In the 2025 SANS survey, almost half of all incidents were detected within 24 hours and 60 % were contained within 48 hours. Even so, full remediation took longer than a month in 19 % of cases. Fast detection, in other words, says little about the actual time to recovery.
Does an OT incident have to be reported to the BSI under NIS2?
Under the NIS2 implementation in the BSIG, operators of critical facilities and regulated entities must report significant security incidents to the BSI within 24 hours, followed by staged follow-up and final reports. A production outage caused by ransomware on OT-adjacent systems typically falls into this category. Meeting the deadline requires a well-rehearsed escalation path, not just a monitoring tool.
Resilience is an exercise, not a product
Industry has poured a lot of money into detection in recent years, and it’s paying off. What the SANS data reveals is the gap behind it. A plant that detects an attack in six hours and takes five weeks to get back to rated capacity doesn’t have a detection problem. It has a recovery problem, and no additional sensor will fix that.
The comparison between regulated and unregulated sites carries the most encouraging message. In 2025, mandatory processes, reporting channels and evidence requirements correlated with almost 50 % lower impact. Paperwork alone doesn’t protect anyone, but processes that are enforced eventually get practiced, and that’s exactly what makes the difference. You can prescribe that practice for yourselves without waiting for the next regulation.
- Change the metric: Time to restored production instead of time to alert.
- Live testing: Emergency operation and restart with a real shift crew, not around a conference table.
- Name your single points of failure: OT vendors, the virtualization platform and remote maintenance belong on the same risk register as the control system itself.
- OT pentesting: Why SCADA, PLCs and industrial plants need a different security logic
- What a breach really costs: Production downtime, IR costs, customer churn and personal liability
A real incident won’t negotiate. Are you ready for it?
We test OT environments under realistic conditions and walk through with you what can really happen in an emergency between detection and restart.
Request an initial consultation →