Start with a hospital that has no internet connection. Then justify every connection you put back.

On 2 November 2026, a cyber incident at a refrigeration supplier disrupted cold-storage controls at three sites of France’s national blood service, EFS. The reported temperature setting was 110 °C. That was a setting entered into the system, not a temperature the cold rooms reached. EFS reported no loss of blood products and no identified harm to patients. The incident came amid attacks on supermarket refrigeration across Europe, following a similar wave in October.

An EFS spokesperson told Reuters that staff responded to alerts and moved supplies to “refrigerators that were not connected”. Alarms, people and an available alternative helped prevent a much worse outcome.

I keep coming back to that alternative. We spend so much time discussing how to secure connected equipment that we can forget to ask why it needs an internet connection in the first place. When the equipment stores plasma for transfusions, that question belongs in the boardroom. Vincent Strubel, director of France’s cybersecurity agency ANSSI, urged operators to disconnect affected Danfoss refrigeration controllers from the internet or apply very restrictive network filtering while the attacks were being investigated.

My starting position is deliberately uncomfortable: take hospitals off the internet. Make disconnection the starting point in their design and require a patient-care case for every exception. Turning that into practice requires clinical planning; abruptly cutting off a working hospital could itself harm patients.

The freezer is already part of patient care

A freezer does not have to touch a patient to be critical to treatment. Neither does a laboratory system. Our assessment of clinical safety has to include the equipment and suppliers on which care depends, even when they sit outside the hospital building.

The 2024 ransomware attack on pathology provider Synnovis illustrates the stakes. The UK government has confirmed that it delayed more than 11,000 appointments and procedures and contributed to a patient’s death. That does not establish that a medical device was hacked. It shows how disruption elsewhere in the care process can have consequences at the bedside.

For an example involving a medical device, look at the Contec CMS8000 and relabeled Epsimed MN-120 patient monitors. In January 2025, the FDA warned of vulnerabilities involving unauthorized remote control and the unauthorized transfer of patient data. Its July 2025 update announced a patch that removed networking functionality altogether, so the affected devices could only be used for local monitoring. The FDA reported no known related incidents, injuries or deaths at the time of that communication.

The Contec case does not tell us that every infusion pump is vulnerable. It does give us a specific example of a safety response that involved removing connectivity. We should be willing to consider that option before an incident forces the conversation.

In 2021, CISA also published an advisory covering remotely exploitable vulnerabilities in the network modules and supporting software of the Fresenius Kabi Agilia Connect infusion system. It concerned specific versions, for which updates were subsequently recommended.

The case for a connection has to be made

Remote monitoring can detect a fault early. A specialist can review a scan from another location. Sharing results can prevent delays. Those benefits deserve a proper assessment, alongside the consequences of compromise or loss of service.

“The supplier requires it” is not enough. I want to know what the connection actually does. Permission to read a temperature is different from permission to change it. Sending an alarm is a different function from allowing remote maintenance. Bundling them together makes purchasing easier, but it also makes it harder to decide which access is justified.

AI gives us another reason to stop relying on technical complexity as protection. In October, Claroty reported reproducing known refrigeration-controller vulnerabilities in a laboratory using Claude Sonnet 4.6, in about 28 minutes at roughly $6 in AI usage. A researcher had to correct and guide the agent. This was a controlled experiment, not evidence that AI caused the French attacks.

My conclusion is that familiar weaknesses may become cheaper to exploit. A hospital should not base a safety decision on the hope that its equipment is too obscure or too difficult to attack.

Design for care without the public internet

An internal clinical network and the public internet are different things. Devices may need to exchange data inside a hospital without exposing their controls online. I want critical clinical functions designed to continue safely when external connections are lost, with the limitations known in advance.

There is already clear guidance on the technical basics. On 5 November, CISA and partner agencies urged operators to remove public internet exposure from refrigeration control systems, separate operational equipment from business networks and secure necessary remote access. Health-ISAC also recommends time-limited supplier access and independent alarms.

My proposal adds a purchasing rule: a supplier must demonstrate what continues to work without its cloud service or remote connection, for how long and with what staffing. If a critical function cannot continue locally, the clinical team must explicitly assess that dependency and the alternative arrangements before the contract is signed.

This also changes what we buy. Independent alarms and safe physical operating limits matter as much as a dashboard. A hospital needs a controlled route for maintenance and updates and a funded plan for replacing unsupported equipment. Keeping something offline does not remove the need to maintain it.

The transition must be tested with clinicians, biomedical engineers, facilities teams and IT. The isolation guidance that the UK NCSC published in January 2026 with partners including the Dutch NCSC explicitly addresses dependencies and data flows that may need to remain available for safety. The purpose is to preserve care when a connection fails or must be closed.

One supplier outage can affect several hospitals

In April 2026, after the ransomware attack on electronic patient record supplier ChipSoft, connections to its Zorgportaal, HiX Mobile and Zorgplatform services were disabled as a precaution, on the advice of Z-CERT. On 16 April, ChipSoft informed affected customers that patient data had been stolen from environments it hosted. Disconnection can contain risk, but it can also leave people with extra work and reduced services.

That is why continuity planning has to extend beyond one building. If several hospitals rely on the same supplier, transferring patients to a neighboring hospital may not be an option. I would ask hospitals in the region to rehearse a shared supplier outage together, with clinical teams deciding in advance what care can continue, what must be deferred and where extra capacity would come from.

Use P-SEP to assess the decision

My P-SEP model brings People, Security, Ethics and Privacy into the same decision. Here, People means giving staff the training, time and authority to act on an alarm, including enough people to deliver the fallback service. Security means knowing the connections and permissions, maintaining the equipment and testing how to isolate a problem.

Ethics means weighing who gains from remote access against who bears the consequences when it fails. Privacy means knowing which patient data leaves a device, where it goes and who can access it. A lower maintenance bill is only part of that calculation.

For the next board meeting, I would ask for evidence on three points:

  • Which external connections support essential care, which merely make administration or maintenance easier and who accepted the risks?
  • What care can continue safely if a key supplier or the internet becomes unavailable, for how long and with how many staff?
  • When did we last rehearse that with the people who deliver the care? Which problems from that exercise remain unresolved?


I want boards to be able to show that they have removed unnecessary dependencies and funded the alternatives. Counting connected devices tells us very little about whether a hospital is becoming a better place to receive care.

RESET! Start with a hospital that has no internet connection. Then make the case for every connection you put back. The burden of proof belongs with the person proposing the connection.

About this article

Sources checked on 6 November 2026. The sources cited here describe ongoing investigations into the refrigeration incidents and do not establish the exact attack route. The policy proposed here is my position; the cited guidance does not call for an indiscriminate shutdown of hospital connectivity.

AI assisted with research, source checking, drafting and editing. The position and final editorial decisions are mine. If you see an error or have relevant evidence to add, please let me know.

Related reading: my article on Albert Heijn, cyber resilience and continuity planning (in Dutch).

Picture of Erik Jan Koedijk

Erik Jan Koedijk

Erik Jan Koedijk is a keynote speaker, cybersecurity expert and AI realist. He shakes organisations awake, makes sense of what is happening and gives perspective.
About Erik Jan

More from this category

AI

‘Girl You Know It’s Real’

Yes, that really is Fab Morvan from Milli Vanilli congratulating me on…

Cybersecurity

AI & Cyber Resilience Masterclasses -Caribbean Tour 2026 – Join the AI Masterclass in Aruba and Curaçao

AI is changing how we work. But is your team ready to…
AI

Reset! The Workplace with AI. The Aruba In House edition 2026!

RESET! Masterclass AI: Aruba edition. Why you should atttend? Discover how to…

Enough reading

Your organisation deserves more than a nice story. Time for action.