Tracking the fridges that went offline
When you think of critical infrastructure, what do you think of? I bet you didn't think of fridges until the supermarket fridge attack in France on October 12th. For those who don't pay attention to the news (to be fair
When you think of critical infrastructure, what do you think of? I bet you didn't think of fridges until the supermarket fridge attack in France on October 12th.
For those who don't pay attention to the news (to be fair, I only knew about this because it got classified as a cyberattack right away), what happened is that fridges in some supermarket chains in France went from 3 Β°C to 50 Β°C. That setting was changed remotely and it appears that it was only detected when it was too late and food got wasted.
So, I got curious. How many fridges got disconnected after the attack? And did traditional perimeter protection measures (such as VPNs) go up in the aftermath?
From 3 Β°C to 50 Β°C... changed remotely

Illustration. The peak was reported by Reuters, in October 2026.
Then, in early November, fridges in other countries in Europe got attacked (including French blood banks).
Europe's fridges, attacked country by country

Dates and countries from press reports: Guardian, Euronews/AFP, AP, Deutsches Γrzteblatt. Map not to scale.
You might ask, why are fridges online? It is a simple way for technicians to remotely monitor and control the temperature. However, these controls are not supposed to be directly accessed from the internet. In theory, they should only be accessed from inside a protected network (in the old "protect the perimeter" approach) or, ideally, via Zero-Trust Network Access, the newer approach to protecting private endpoints.
Most people don't even think that fridges and other pieces of critical equipment might be "online" the way our computers are. Guess how many fridge controls in Europe could anyone reach online, out in the open?

Data: The Shadowserver Foundation, IoT device statistics, vendors Danfoss, Copeland and Carel, Europe, 11 Oct 2026.
5,985 fridge controls online (and those are "just" the ones that Shadowserver Foundation detected on their IoT device statistics) is 5,985 fridge controls too many to be out in the open (by the way, I'm also wondering why fridges can accept configurations of temperatures up to 50 Β°C, but I might be digressing here... or not. Validation error?).
Every dot below is one fridge control anyone could reach directly

Data: The Shadowserver Foundation, 11 Oct vs 11 Nov 2026. Vendor fingerprints may include some non-fridge devices from the same makers.
What happened in the aftermath? I thought that pretty much all of these fridges would disappear from the charts. In fact, France, Belgium and Romania reacted fast to the attacks and disconnected the fridges (or protected them). Poland followed more slowly, Italy only partly, and Spain barely moved.
Who unplugged after the attacks?

Data: The Shadowserver Foundation, daily, late SepβNov 2026; countries with 150+ controls on 11 Oct.
Then I started wondering... Did the "old perimeter" (VPNs and firewalls) go up? Did controls for other types of industrial equipment get affected too?
So, what did the attacks move after all?

Data: The Shadowserver Foundation, Europe, daily, late SepβNov 2026; 39 vendors grouped by what they mostly make. A vendor isn't a category, so the groups are rough.
What I looked for, and what I found
| Guessed | Checked | Found |
|---|---|---|
| Shops unplugged their fridge controls after the attacks | Daily counts of Danfoss, Copeland and Carel controls reachable online in Europe (Shadowserver), 11 Oct vs 11 Nov. | 5,985 β 2,550 (β57%). Yes |
| Some countries reacted much more than others | The same counts, country by country | Romania β92%, Belgium β86%, France β77%, Italy β33%, Spain β2%. Yes |
| It was a wider clean-up, not just fridges | 39 vendors in six groups: building controls, VPNs and firewalls, solar and power, 4G routers, industrial controllers | Everything else moved between β1% and β5%. Only the fridges fell. No |
| They put the fridges behind a classic VPN or firewall | VPN and firewall login pages online in Europe; job ads that mention VPNs in EU retail, 32 days before vs after 12 Oct | Login pages β5%. VPN job ads flat (45 β 54, the same share of all ads). No sign |
| They moved to zero-trust access (ZTNA: Zscaler, Cloudflare, Netskope, Barracudaβ¦) | EU retail job ads that mention ZTNA/SASE products; public DNS records of 49 large grocery chains in 14 countries; certificate logs for new remote-access hostnames | Job ads 8 β 6. DNS: 1 chain of 49 shows a ZTNA portal, and DNS can't say since when. Certificate logs (public logs reach only 23 Oct): 3 new remote-access-style names in the 11 days after the attack, none of them ZTNA, against 29 in the 32 days before. No sign |
| Retailers started hiring for factory-style (OT) security | EU retail job ads that mention OT or industrial-control security | 7 β 16. Too few to call a trend. Too few to tell |
ZTNA is built to be invisible from outside, and job ads lag decisions by months, so maybe too early to tell.
Sources: The Shadowserver Foundation; job ads from TheirStack, searched with looot; public DNS records and certificate-transparency logs (crt.sh). Counts only: no company is named.
So, it appears that only the fridges moved. Everything else stayed within a few percent of what was observed before, and there's no public sign of what replaced the open connections. Of course, they could have moved to inside the existing VPNs (how did they end up exposed on the internet?). And zero-trust access is, rightly so, invisible from the outside by design, so we have no way of knowing whether these devices were disconnected or moved to a protection layer.
Why did the blood banks manage to stop the attack in time? Because monitoring (and perhaps alerts) still worked. But unplugging is the emergency brake, and, again, from the outside we can't tell whether those 3,435 controls were unplugged or suddenly protected. If they were unplugged, their remote monitoring might have gone dark too.
At home, the hubs stayed online

Data: The Shadowserver Foundation, Europe, daily, late SepβNov 2026. Smart-home hubs = device type smart-home-management (Home Assistant, Somfy, Philipsβ¦). Gaps are days the scan missed. No home fridge maker shows up in the scans.
One wonders, what about our fridge at home? I couldn't find home fridge makers in Shadowserver's scans. The "home appliance" group in Europe included 7 devices on October 11th, and they were all robot vacuums.
The reason is that the architecture itself is different: home smart fridges (such as Samsung, LG and Bosch) connect directly to the maker's cloud and then you control them from an app. So the way "into your fridge", besides the shopping cart, is your account. This already happened in 2017, when a bug in LG's SmartThinQ app let attackers take over accounts and the appliances in them.
There wasn't much change in the number of hubs online in the aftermath of the attack: a month after the attacks, 95% were still online, against 43% of the shop fridge controls. That being said, there might be a door to other devices at home, since a scan can see the 84,767 smart-home hubs on October 11th, 73,313 of them Home Assistant. A reachable hub still asks for a login, but it's worth it to consider using zero-trust access or a VPN in order to avoid having problems if there are bugs in the Home Assistant software.
What lessons can we take from this?
I'm definitely a fan of automating tasks that shouldn't be performed by humans. But we often add the remote access and not the right level of protection. We have been listening to the "security is everyone's responsibility" motto for years now... maybe now is the time to start acting on it.
I like to think security-first. And anything that gets connected to a system needs to have a tight security control with appropriate observability. Including (you guessed it!) AI agents.
Method: counts from the Shadowserver Foundation's public IoT dashboard (Europe, daily, by country), vendors Danfoss, Copeland and Carel, 11 Oct vs 11 Nov 2026; 39 vendors in six groups for comparison. Job ads from TheirStack (EU retail, the 32 days before vs after 12 Oct). Public DNS records and certificate-transparency logs (crt.sh) of 49 large grocery chains in 14 countries: counts only, no company named. News and job ads found with looot (disclosure: I'm a cofounder).
Caveats: a vendor match can include other devices from the same makers; a control going offline can mean unplugged, firewalled, moved behind a VPN or zero-trust access, or just not answering that day; job ads lag decisions by months; the public certificate logs only reach 23 Oct.
Data: The Shadowserver Foundation (dashboard). Sources: Reuters, 12 Oct; Guardian, 2 Nov; Euronews, 3 Nov; AP, blood banks; Deutsches Γrzteblatt.
Originally published on luisalima.com.
Originally published by Dev.to Security. Aggregated on AIWithGhost for educational purposes β full credit and traffic to the original publisher.