loader image

Home → Blog →

The Edge Box in Hall 2 Hasn’t Had an Update Since Commis­sioning

“too long,
didn’t read”

1

If your plant has run a project that brings machine data to a dash­board, an MES, or the cloud, check whether a small indus­trial PC in a control cabinet collects and forwards that data. In the scene this post starts with, an inte­grator installed such a box, it has run without failure since, and nobody has logged in since the accep­tance test. The box is a full computer between the machine network and other networks, and its soft­ware stays at the version it had on the day of commis­sioning while new vulner­a­bil­i­ties keep being published. Uptime tells you the box works. It tells you nothing about whether its soft­ware is up to date.

2

In that scene, the problem is orga­ni­za­tional, not tech­nical. The inte­gra­tor’s contract has ended, the machine builder supports the machine, plant IT was never told about the box, and main­te­nance is respon­sible for avail­ability, not for oper­ating system updates. The first step is there­fore an inven­tory: every edge device, what it runs, who installed it, and one person respon­sible for its updates. Without that list, nobody can say which devices need an update.

3

The EU Cyber Resilience Act sets cyber­se­cu­rity require­ments for manu­fac­turers of prod­ucts with digital elements. Its reporting duties apply since September 11, 2026, its remaining oblig­a­tions from December 11, 2027, and it contains no oblig­a­tion to retrofit a device you bought years ago. What you can do your­self is a monthly update routine planned around produc­tion, docu­mented protec­tive measures where no update exists, and four purchasing ques­tions for the next device. For shared, vendor-neutral device manage­ment, the OI4 Alliance has work­groups for Remote Device Manage­ment, Indus­trial Cyber Secu­rity, and Open Edge Computing.

Uptime is not
a patch level

Lucas Wolf, Tech­nical Content Manager @Open Industry 4.0 Alliance

November 6th

THE GRAY BOX ON THE DIN RAIL.

Picture a scene. Hall 2, an injec­tion molding line with six machines. In the control cabinet next to the last machine, between the 24-volt power supply and a managed switch, sits a fanless gray box about the size of a paper­back. A label maker sticker on the front says “EDGE-02. Do not switch off.” Inside the cabinet door, in a plastic sleeve, there is a printout from 2019 with an IP address, a user name, and a pass­word that someone under­lined twice.

The box reads cycle times and barrel temper­a­tures from the six machines over OPC UA and sends them via MQTT to the dash­board the plant manager opens every morning. It has done this reli­ably for years. It was part of a digi­tal­iza­tion project: the inte­grator deliv­ered, the accep­tance protocol was signed, the project folder was archived.

Since then, nobody has logged in. Not because anyone decided against it. There was never a reason. The dash­board works.

This scene is invented. The ques­tion it raises is real: does your plant have a box like EDGE-02, and who looks after it?

RUNNING IS NOT THE SAME AS MAIN­TAINED.

An edge device is an indus­trial computer close to the machine that collects data, preprocesses it, and forwards it to another system. Depending on the product, it runs a general-purpose oper­ating system such as Linux, plus compo­nents like an OPC UA client, an MQTT client, a container runtime, a web inter­face for config­u­ra­tion, and a remote access compo­nent. Each of these is soft­ware. For each of them, new vulner­a­bil­i­ties can be published after the day of commis­sioning.

In the scene, the controllers of the six machines sit in the machine network. EDGE-02 connects two networks: one port toward the machines, one toward the plant network, through which the data reaches the dash­board. Connecting networks is its purpose. It also means that anyone who gains access to the box has a network path into the machine network, which is why OT secu­rity has to include edge devices.

The soft­ware on EDGE-02 does not change. The list of known vulner­a­bil­i­ties does. Every vulner­a­bility published since commis­sioning that affects the installed versions stays open until someone installs an update. For manage­ment, this does not show up in the usual reports: the box has not failed once, so it appears in no OEE report and on no main­te­nance plan.

NOBODY’S NAME IS ON IT.

Back in the scene: ask around the plant who is respon­sible for updating EDGE-02, and you get five answers, each of them correct from its own point of view.

The inte­grator deliv­ered a project. The contract covered instal­la­tion and accep­tance, not oper­a­tion, and the engi­neer who config­ured the box now works for another company. The machine builder supports the six injec­tion molding machines, not a third-party box in the cabinet next to them. Plant IT manages laptops, servers, and the office network and was not involved in the project, so the box is not in its asset list. Main­te­nance is respon­sible for avail­ability. It knows the box is there, but updating a Linux system is not part of its job descrip­tion. Manage­ment signed off the digi­tal­iza­tion project as completed, which it is.

Five correct answers add up to one box that nobody owns. You can test this in your own plant with one ques­tion per device: who would install an update next Tuesday, and with which login? If nobody can name a person and a login, no update will be installed. The missing piece is not a patch. It is a name.

FIRST, COUNT THEM.

You cannot update what you have not found. The IEC tech­nical report on patch manage­ment for indus­trial systems, IEC TR 62443–2‑3, lists an inven­tory of the existing envi­ron­ment as part of setting up a patch manage­ment program. Four sources help you build that inven­tory.

Start with the project folders. Every project that brought machine data to a dash­board, an MES, or the cloud needed a compo­nent that collects the data, so check for each one whether that compo­nent is a sepa­rate device. Purchase orders and accep­tance proto­cols also name the supplier, which a network scan cannot tell you.

Then look at the switches. The address tables of managed switches list every device that commu­ni­cates through them, and the first half of a MAC address iden­ti­fies the manu­fac­turer of the network inter­face. Do this together with IT. An active scan sends requests to every device on the network, so agree with your machine builders before­hand whether their controllers tolerate it in produc­tion.

 

At the next planned shut­down, open the cabi­nets. Main­te­nance walks through with a flash­light and a phone camera, and a photo of the type plate is enough for now.

Finally, ask the compa­nies that built your lines. Inte­gra­tors and machine builders can tell you which computers they installed and under which contract.

For each device, write down the same seven things: loca­tion, manu­fac­turer and model, oper­ating system or firmware version if you can read it, what it connects to (machines, networks, cloud endpoints), who installed it, whether a support contract or announced end of support exists, and one person respon­sible for updates.

The last column is the one that matters. A spread­sheet with forty rows and no names is an inven­tory. The same spread­sheet with forty names is a plan.

AN UPDATE ROUTINE THAT FITS A SHIFT PLAN.

The same IEC report describes patch manage­ment as a set of processes: watching for releases, deciding which patch goes to which system, deciding whether to test first, choosing when to install, and tracking that it worked. For a plant with a few dozen edge devices, that trans­lates into five steps.

First, watch. For every manu­fac­turer on your list, find out where they publish secu­rity advi­sories, and let one named person check them on a fixed day each month.

Deciding comes next, because not every patch is urgent. A flaw in a web inter­face that is reach­able only from a locked main­te­nance network is not the same as a flaw in the remote access compo­nent that faces the internet.

Before the first update, take a full image of the box and store the config­u­ra­tion export next to it. If you have an iden­tical spare, test there. If not, start with the least crit­ical line.

Install in a window you already have: planned main­te­nance, a shift change, a weekend. If a box only reads and forwards data, a restart inter­rupts the data stream to the dash­board. Check per box whether anything in produc­tion depends on it, for example setpoints it writes back to a machine. If so, plan the restart like any other inter­ven­tion in the process.

After­wards, record the date, the version before and after, and who did it. Three more columns in the same spread­sheet.

Some­times there is no update. The vendor no longer exists, the hard­ware has reached its announced end of life, or the box is part of a vali­dated system you cannot change without requal­i­fi­ca­tion. Then docu­ment what the box can reach, close services nobody uses, switch off remote access until someone needs it, replace default pass­words, and put the device on the replace­ment list with a date. “We cannot patch this, so we restricted its connec­tions and replace it in the third quarter” is a docu­mented deci­sion with an owner and a dead­line.

WHAT THE CYBER RESILIENCE ACT WILL AND WON’T DO.

The Cyber Resilience Act, Regu­la­tion (EU) 2024/2847, sets cyber­se­cu­rity require­ments for prod­ucts with digital elements made avail­able on the EU market. It entered into force on December 10, 2024. Since September 11, 2026, manu­fac­turers must report actively exploited vulner­a­bil­i­ties and severe inci­dents, including for prod­ucts already on the market. The essen­tial require­ments and the remaining manu­fac­turer oblig­a­tions apply from December 11, 2027. For an intro­duc­tion, see our webinar on the EU Cyber Resilience Act.

One of those oblig­a­tions is the support period, the time during which a manu­fac­turer must provide secu­rity updates. As a rule it is at least five years, shorter only if the product is expected to be used for less time. The Commis­sion’s guid­ance from July 2026 adds that five years is not a universal stan­dard: a product with a longer expected life may need longer support. The end date has to be commu­ni­cated to users.

Who is affected? Manu­fac­turers of edge devices are addressed as manu­fac­turers of prod­ucts with digital elements. Whether a machine builder is addressed for a machine depends on whether the machine counts as a product with digital elements, and inte­gra­tors who combine hard­ware with their own soft­ware and deliver it as a product should clarify whether they take on manu­fac­turer duties. Both are ques­tions for legal counsel. How remote main­te­nance, soft­ware updates, and audit trails were imple­mented for CRA compli­ance in an indus­trial project was the topic of our webinar CRA in Prac­tice. Oper­a­tors that only use a device are not among the parties the CRA places oblig­a­tions on, and the regu­la­tion contains no oblig­a­tion to bring a device from 2019 up to its new require­ments. The inven­tory and the names stay your job.

From the CRA require­ments you can derive four ques­tions for your next purchase. How are updates deliv­ered, and can our own staff install them without a service visit? When does the support period end, and where is that written? Where do you publish secu­rity advi­sories? Can we manage the device with a tool of our choice, or only through your portal?

SIX VENDORS, SIX PORTALS.

The last purchasing ques­tion is the one a single plant cannot answer for itself. Forty edge devices from six vendors can mean six manage­ment portals, six logins, six update proce­dures, and six ways to set up remote access. Every addi­tional tool adds a login, a proce­dure, and a person who needs to know both.

One tool for devices from several vendors requires common inter­faces for onboarding, moni­toring, and updating that every manu­fac­turer supports. No oper­ator can define them alone, and no single vendor can make the others adopt them. That is pre-compet­i­tive work. In the OI4 Alliance, Open Edge Computing compli­ance defines require­ments for inter­op­er­able edge solu­tions, and our hackathon on public key infra­struc­tures worked on edge secu­rity, secure MQTT commu­ni­ca­tion, and certifi­cate life­cycle manage­ment.

BACK TO HALL 2.

Imagine EDGE-02 six months from now. The printout in the cabinet door is gone, and the pass­word lives in the plant’s pass­word manager. The sticker on the front has a second line: an owner’s name and the date of the next check. The box still feeds the dash­board every morning. From the outside, nothing has changed. That is the point.

You can start tomorrow after­noon without a budget. Take one hall, one flash­light, and a spread­sheet with the seven columns from above. Count the edge devices and fill in the last column. Every row without a name goes on the agenda of your next meeting with IT and main­te­nance.

Making forty devices from six vendors manage­able in one way is the part no plant solves on its own. This is what the Open Industry 4.0 Alliance is for. Oper­a­tors, machine builders, tech­nology providers, and inte­gra­tors work together there in the pre-compet­i­tive space, on ques­tions like the ones in this post: which inter­faces vendors should support, how updates and remote access can work across manu­fac­turers, and what an oper­ator can require when buying.

The work happens in projects with a defined goal and in demon­stra­tors, where members show on real hard­ware that devices from different vendors work together. Around that sits a network of experts from oper­a­tors, machine builders, and tech­nology providers. Bring your inven­tory or your purchasing ques­tions, discuss them with people who deal with the same ques­tions in their own compa­nies, and see which solu­tions others have already tested. You can take part as a project partner without being a member.