Key takeaways
- Safety first: the shift supervisor and process safety decide on stopping or isolating, not the SOC alone.
- Every critical line needs a documented manual or degraded mode and the people trained to run it.
- AI failures belong in the plan: a poisoned or drifting model can cause a quality incident without any malware.
- Under NIS2 Article 23, in-scope entities give an early warning within 24 hours, an incident notification within 72 hours and a final report within one month of the notification.
- Run tabletops on realistic scenarios: ransomware on MES, a degraded QC model and a compromised OEM remote session.
When ransomware hits a manufacturer, the first hard question is rarely technical. It is "do we keep the lines running?" The answer depends on safety, on whether you can operate without the affected systems and on who has the authority to decide. A good OT incident response plan answers those questions before the alarm goes off. This guide covers the plan, the playbooks, AI-specific failure modes, NIS2 reporting and three tabletop scenarios to test it.
What makes OT incident response different
| Topic | Typical IT response | OT response |
|---|---|---|
| First priority | Contain and protect data | Protect people, equipment and process state |
| Isolation | Quarantine endpoints quickly | Decide with operations; isolating can stop the process unsafely |
| Recovery | Reimage from gold image | Restore PLC programs, HMI projects, configs from verified backups; vendor may be needed |
| Evidence | Disk and memory images | Also controller logic comparisons, historian data, network captures |
| Decision maker | CISO / incident manager | Plant manager or shift supervisor for process, with incident manager |
| Fallback | Workarounds, cloud services | Manual operation, degraded mode, controlled shutdown |
NIST SP 800-82 Rev. 3 covers incident response for OT in this spirit. The key point for a mid-market plant: the IT plan alone is not enough. You need an OT annex with plant decision rights and process-specific steps.
The plan: what to write down
Keep the core plan short. Detail goes in playbooks.
- Roles and decision rights. Incident manager, plant lead, OT security lead, process safety, communications, legal, and the SOC or MSSP. Who can order a line stop, an IT/OT disconnect or a plant shutdown. See IT/OT convergence governance.
- Call tree. 24/7 contacts for each site, key OEMs and integrators, your incident response retainer, insurer, and the national CSIRT if NIS2 applies.
- Severity levels. Tied to production, safety and quality impact, not only to IT categories.
- Isolation points. Pre-agreed: the IT/OT firewall, the DMZ, the remote access broker, line-level boundaries. Know what each one breaks before you pull it. Your zones and conduits drawings are the map.
- Manual and degraded modes. For each critical line: can it run without MES, without the historian, without the HMI, without the AI system? For how long? Who is trained?
- Backups and recovery. Offline, tested copies of PLC programs, HMI projects, SCADA configs, firewall configs and model versions. Know which version is known-good.
- Reporting. Regulatory, customer and insurer deadlines, with who drafts and who signs.
Plant-floor reality: Many plants discover in the middle of an incident that nobody under 40 has run the line manually, and that the paper travelers were thrown out years ago. Check now. Run a planned manual-mode drill during a slow shift.
Safety first, always
- The safety instrumented system must remain independent of the systems you are investigating. Never use it as a containment tool.
- If you cannot trust what the HMI shows, treat the process as unverified. Use local indications and field checks, and move toward a safe state.
- Containment actions in OT are approved by the plant lead with process safety input. The SOC recommends; operations decides.
- Log every manual intervention for later reconstruction.
AI systems in the incident plan
AI adds two things to plan for: the AI system as a victim or entry point, and the AI system as a cause.
The December 2025 joint guidance from CISA and partner agencies, Principles for the Secure Integration of Artificial Intelligence in Operational Technology, recommends including AI in incident response plans, defining thresholds that switch back to non-AI operation, and adding procedures for bypassing or replacing AI to functional safety and incident response plans.
For each AI use case in your AI use-case register, document:
- Fallback mode: what runs when the model is off. For example, fixed recipe, manual inspection, time-based maintenance.
- Trigger thresholds: accuracy, false reject rate or confidence levels that force fallback.
- Known-good model version and how to roll back.
- Data dependencies: which sensors and systems feed it, so you can check for tampering or drift.
- Owner: who decides it is safe to switch back on.
See securing AI in OT networks for the preventive controls.
NIS2 reporting timelines
If your company is an essential or important entity under NIS2 (the Directive's annexes list manufacturing of certain products, such as machinery, motor vehicles, electrical equipment and electronics, as well as the chemicals and food sectors), Article 23 sets the reporting cascade for significant incidents:
| Deadline | Report | Content (summary) |
|---|---|---|
| Within 24 hours of becoming aware | Early warning | Whether malicious action is suspected and whether there could be cross-border impact |
| Within 72 hours of becoming aware | Incident notification | Update of the early warning, initial assessment of severity and impact, indicators of compromise where available |
| On request | Intermediate report | Status updates requested by the CSIRT or competent authority |
| Within one month of the incident notification | Final report | Detailed description, likely root cause, mitigations applied, cross-border impact |
If the incident is still ongoing at the one-month mark, you provide a progress report then and a final report within one month of handling the incident.
As of November 2026, NIS2 transposition is still uneven. In mid-2026 the European Commission referred several member states to the Court of Justice for failing to transpose on time. Check the national law in each country where you operate, including which authority receives reports. See NIS2 for manufacturers.
Note: If you also make products with digital elements, the Cyber Resilience Act's reporting obligations for actively exploited vulnerabilities and severe incidents have applied since 11 September 2026. That is a separate duty from NIS2. See the Cyber Resilience Act.
The 24-hour clock is the hard one. Pre-draft an early warning template and agree who can send it at 2 a.m. on a Sunday.
Playbooks
Write short playbooks, one or two pages each, for the scenarios most likely to hit you:
- Ransomware in IT spreading toward OT
- Ransomware or wiper on OT Windows systems (HMI, SCADA, MES)
- Unauthorized change to controller logic
- Compromised vendor remote access
- AI model degradation or suspected tampering
- Loss of view (HMI or SCADA unavailable)
Each playbook: indicators, first 30 minutes, isolation options and their production impact, who decides, fallback mode, recovery steps, evidence to preserve, reporting triggers.
Tabletop scenarios
Run these with IT, OT, operations, process safety, quality and a management sponsor. Ninety minutes each is enough.
Scenario 1: ransomware hits the MES
Setup: Monday 06:00. The MES server and two engineering workstations at Level 3 show ransom notes. Lines are still running on their last loaded orders. The SOC sees the same strain on IT file servers.
Questions to test:
- Do you cut the IT/OT firewall now? What breaks when you do?
- Can lines run without MES? For how many hours? Who issues work orders and records genealogy by hand?
- Are PLC programs and HMI projects backed up offline, and when were they last verified?
- Who decides to restart MES, and on what evidence?
- Does this meet the NIS2 significant incident threshold? Who sends the 24-hour early warning?
Scenario 2: a poisoned or drifting model degrades QC
Setup: Over three weeks, customer returns for a surface defect rise. The vision inspection model on assembly has been passing parts it used to reject. Investigation finds the model was retrained last month on a dataset where a batch of defect images was labeled as good. It is unclear whether this was a mistake or deliberate.
Questions to test:
- How fast could you detect this? What monitoring exists on false accept rate?
- What is the fallback? Is manual inspection staffed and trained?
- Who owns the decision to switch the model off, and who decides it is safe to switch back on?
- Can you identify the affected lots and notify customers?
- How do you tell accident from tampering? Who reviews training data access logs?
- At what point does this become a security incident, and is it reportable?
Scenario 3: OEM remote session compromised
Setup: The furnace OEM calls. One of their engineers' laptops was compromised and their credentials used. Your broker logs show a session to the heat treatment HMI at 02:14 last night outside any approved window. A recipe parameter was changed.
Questions to test:
- Could this session even happen? Was access just-in-time, or always-on? See secure remote access.
- How do you verify the current recipe and PLC logic against known-good versions?
- Is product from last night's run affected? Who checks?
- Do you disable all OEM access, or only this vendor's?
- What does the contract say about the OEM's duty to notify you? Check it with the OT/AI vendor security questionnaire answers.
What good looks like
- An OT annex to the incident response plan with plant decision rights and a call tree.
- Tested manual or degraded modes for each critical line.
- Offline, verified backups of controller programs, configurations and AI models.
- AI fallback modes and thresholds documented per use case.
- A pre-drafted NIS2 early warning template, if in scope.
- At least one OT tabletop per year, with tracked actions.
Effort and cost
Writing the OT annex and the first four playbooks takes a few weeks of part-time effort from the OT security lead with input from operations. Each tabletop costs a couple of hours of senior time plus preparation. An incident response retainer with OT capability is a recurring cost worth comparing against the cost of one day of plant downtime.
Common mistakes
- An IT-only plan that tells the SOC to isolate endpoints, including HMIs.
- Backups of PLC programs that live on the same network as the systems they protect.
- No plan for running without the MES.
- Treating an AI quality failure as "a data science problem" with no incident owner.
- Discovering the NIS2 reporting contact on day two.
Monday-morning checklist
- Confirm who can order a line stop and an IT/OT disconnect at each site. Write it down.
- Check when PLC and HMI backups were last taken and whether one is offline.
- For each critical line, write one paragraph on how it runs without MES.
- Add fallback mode and owner to every AI use case touching OT.
- Confirm whether NIS2 applies and draft an early warning template.
- Book a 90-minute tabletop on scenario 1 within the next month.
This article is general guidance, not legal advice. Confirm your NIS2 and CRA obligations with counsel.
Sources and further reading
- Directive (EU) 2022/2555 (NIS2), EUR-Lex
- NIST SP 800-82 Rev. 3, Guide to Operational Technology (OT) Security
- Principles for the Secure Integration of Artificial Intelligence in Operational Technology (CISA and partners, Dec 2025)
- Secure connectivity principles for operational technology (ASD's ACSC, NCSC-UK and partners, Jan 2026)
- Safer and more secure digital products: CRA reporting obligations (European Commission, Sept 2026)
This guide is general information for manufacturers, not legal advice. Confirm regulatory obligations with qualified counsel.
FAQ
Frequently asked questions
How is OT incident response different from IT incident response?
In IT, the default is often to isolate and reimage quickly. In OT, isolating or rebooting the wrong device can stop a process unsafely, scrap a batch or damage equipment. OT response puts safety and process state first, involves operations and process safety in every containment decision, relies on known-good backups of controller programs and configurations, and plans for manual operation while systems are restored.
What are the NIS2 incident reporting deadlines?
For significant incidents, Article 23 of NIS2 requires an early warning without undue delay and within 24 hours of becoming aware, an incident notification within 72 hours, an intermediate report on request, and a final report no later than one month after the incident notification. Reports go to the national CSIRT or competent authority. National laws set the details, so check the transposition in each country where you operate.
Should AI model failures be treated as security incidents?
Treat them as incidents with a defined owner, whether or not malice is involved. A model that starts passing defective parts or giving wrong maintenance advice needs the same steps: detect, contain by switching to fallback, investigate the cause (drift, bad data, tampering), recover with a validated model and learn. If tampering is suspected, it becomes a security incident and may be reportable.
How often should a manufacturer run OT tabletop exercises?
At least once a year with IT, OT, operations, process safety and management, and after any major change such as a new MES, a new remote access platform or a new AI system on a line. Short focused exercises of 90 minutes to two hours with one scenario work better than a full-day event. Track the actions and close them before the next one.
Related guides
Frameworks
NIS2 for manufacturers: scope, duties and what to do first
NIS2 (Directive (EU) 2022/2555) brings many medium and large manufacturers into EU cybersecurity law for the first time.
Guide · 8 min read
OT Security
Securing AI in OT networks: a practical guide for manufacturers
Treat every AI system that touches OT as a new conduit and a new piece of software you have to patch.
Guide · 8 min read
OT Security
Secure remote access for OT: managing OEM and integrator connections
OEMs and integrators need remote access, and it is one of the most common ways attackers reach OT.
Guide · 7 min read
OT Security
IT/OT convergence governance: who owns what in a mid-market plant
IT/OT convergence fails on ownership, not technology.
Guide · 7 min read