The largest implementation mistake would be to treat Annex III §§1.1.9 and 1.2.1 as a cybersecurity feature to be added immediately before CE marking. The requirements touch architecture, component selection, development, integration, commissioning, maintenance and future updates. IEC 62443 reinforces this lifecycle view through the distinct responsibilities of product suppliers, system integrators/service providers and asset owners.
FAT/SAT and commissioning
FAT and SAT therefore need a cybersecurity extension. Traditional FAT proves that the machine performs and that its safety functions operate under designed test conditions. For connected machinery subject to Regulation (EU) 2023/1230, acceptance should also generate evidence that relevant digital attack or failure paths do not invalidate those functions.
FAT/SAT cyber evidence should not be viewed merely as an IT appendix. Where the manufacturer relies on cybersecurity controls to satisfy an applicable Annex III requirement, those controls become part of the reasoning by which conformity is demonstrated. The corresponding evidence should therefore be traceable to the machine risk assessment, architecture, software and configuration baseline, verification activities and technical documentation. Annex IV’s ability to reach safety-related software or programming logic reinforces the point: software state and digital configuration now belong inside the machinery conformity domain when they are relevant to safety.
The practical consequence is that FAT and SAT should establish more than a momentary pass/fail result. They should establish an assurance baseline against which later maintenance, updates and modifications can be compared.
That baseline may include the validated PLC and safety-PLC applications, firmware versions, safety parameters, controller configuration, network interfaces, approved communication paths, remote-access configuration, privileged-account model, software hashes or project checksums where available, firewall rules, backup sets and the evidence generated during functional-safety and cybersecurity verification.
The key distinction is between testing the machine and establishing the state of the machine that was tested. A successful FAT has limited evidential value if the organisation cannot later determine whether the software, parameters, network exposure or access paths operating in production are still the same ones that were validated.
A useful assurance model therefore has four evidence classes:
The same reasoning must then survive commissioning. The lifecycle is not a simple sequence from design to decommissioning; it is a set of assurance gates and feedback loops in which every change must be classified according to whether it preserves the validated baseline, modifies a safety-relevant dependency, or potentially changes the machinery risk model itself.
The central object in this model is the accepted digital baseline created at commissioning. It represents the software and configuration state for which the safety and cybersecurity assumptions were actually validated. From that point onward, operational change management is not merely an ITSM process; it becomes part of preserving the machine’s conformity rationale whenever the changed element participates in a safety function or in the protection of that function.
A useful way to think about the lifecycle is therefore:
validated architecture → verified implementation → accepted baseline → controlled change → impact assessment → revalidation where necessary → updated baseline
rather than:
FAT passed → machine commissioned → maintenance proceeds independently.
This distinction is particularly important for software and configuration changes because the physical machine may appear unchanged while its behaviour, attack surface or safety assumptions have materially changed.
A security patch, for example, may leave the safety envelope completely unchanged and merely restore the integrity of a component. A firmware upgrade may change communications, diagnostics or controller behaviour while leaving the intended safety function intact. A PLC parameter change may alter production performance without affecting risk reduction. Conversely, a seemingly minor modification to a safety limit, interlock mapping, operating mode, communication path or remote-access capability may change an assumption on which the original risk assessment relied.
The lifecycle process should therefore classify changes according to their effect on the assurance case, not according to their apparent size. A one-line change to PLC logic can be more consequential than replacement of an entire HMI. A firmware update affecting only diagnostics may be less significant than opening a previously unavailable remote programming path. The relevant question is whether the change alters a digital dependency that participates in preventing hazardous behaviour.
This is also where cybersecurity vulnerability management and machinery change management converge. A vulnerability may require a patch; the patch changes software; the changed software may participate in a safety-related function; and the organisation must then determine how much verification is required before the machine can return to service.
That does not mean that every security update requires a new conformity assessment. It means that the change process should answer a sequence of progressively stronger questions:
Did the change alter the validated digital baseline?
If not, ordinary maintenance evidence may be sufficient.
Does the changed element participate in, protect, communicate with or influence a safety-relevant function?
If yes, the safety and cybersecurity assumptions associated with that dependency should be reviewed.
Does the change modify the machine’s hazardous behaviour, safety limits, protective functions, control authority or risk-reduction measures?
If yes, revalidation of the affected safety case becomes necessary.
Does the modification meet the complete legal test for substantial modification under Article 3(16)?
Only at that stage does the specific regulatory consequence associated with substantial modification have to be considered.
This layered logic prevents two opposite errors. The first is under-classification: treating every software change as routine IT maintenance even when it changes the machine’s safety behaviour. The second is over-classification: treating every cybersecurity patch or firmware update as though it automatically created a new machine requiring full conformity reassessment.
The correct unit of analysis is the safety-relevant delta between the previously accepted baseline and the proposed state. That delta should be supported by evidence. Depending on the change, the evidence may include a software-version comparison, configuration diff, firmware release assessment, modified network-flow matrix, updated threat model, safety-parameter comparison, regression test, functional-safety validation, remote-access test or restored baseline hash. The purpose is not to create documentation for its own sake; it is to demonstrate that the assumptions used to justify safe operation remain true after the intervention.
This makes configuration management a central part of cyber-physical safety. In a conventional IT environment, configuration management primarily supports service reliability and recovery. In machinery, the same mechanism can also preserve evidence that the logic and parameters enforcing a physical safety constraint are still the ones that were validated.
For the same reason, backup and restore should be understood as more than availability controls. Restoring an old PLC project, safety configuration or controller image can itself create a hazardous state if the restored baseline is incompatible with subsequent mechanical, electrical or software changes. A known-good backup therefore means not only malware-free or technically recoverable, but known to correspond to a valid machine configuration.
The assurance record should consequently follow the machine throughout its useful life. It should connect design intent, verification evidence, commissioned baseline, interventions and subsequent revalidation sufficiently well that a competent person can reconstruct the state of the machine and understand why it was considered safe at a particular point in time.
This is the deeper lifecycle implication of the Machinery Regulation’s treatment of software, intervention evidence and digital modification: conformity is established at a point in time, but the assumptions on which conformity depends can subsequently be changed by software and configuration. Cyber-physical safety therefore requires a mechanism for detecting, assessing and evidencing those changes throughout the operational lifecycle.