Research · Cold chain · 22 November 2026 · 12 min read
One controller, one store: how the European refrigeration attacks worked
Roughly 800 shops in eight countries lost their cold chain to a web login. The way in had been published nine weeks earlier.
Written with AI at the editor's request, from the public sources listed at the end, and checked against them. Each chart shows figures published by the source named under it.
Between 11 October and 2 November 2026, someone raised the temperature in the fridges and freezers of roughly 800 shops in at least eight European countries, and in the cold rooms of a French blood service. No malware was needed and no stolen password has been reported. The attackers logged in to one box per site, the store's supervisory refrigeration controller, over the open internet, and used its own menus.
Key findings
- The supervisory controller is a single point of failure for a whole store. One web login changes the setpoints and defrost schedules of every case and cold room behind it. The attackers never had to touch the plant itself.
- The way in was published months earlier. Claroty's Team82 described login bypasses in Danfoss AK-SM 800A and Copeland XWEB Pro on 9 August 2026. Fixes existed for both. In each product an administrator password could be worked out from the date and details the device gives away.
- The attack used the machine's normal functions against it. Defrost was switched on or stretched from one hour to as many as eight, and targets were set to 50°C. Long defrost delays the temperature alarm, which is why stores found out in the morning.
- It scaled because the sites were the same, not because the attackers were skilled. The same few products, reached the same way by the same contractors, sat on public addresses. More than half of the exposed Danfoss panels counted before the first wave were in France.
- The second wave moved to where the first had not forced a clean-up. French exposure fell by about 70% in three days. Three weeks later the same technique hit six other countries, British food suppliers and a blood service, with many smaller operators among the victims.
France has blamed Iran's Revolutionary Guard. None of the published technical sources shows the evidence for that, and the published record does not confirm which flaw was used at any victim.
Timeline
The warnings came first, by more than a year in one case. The attack followed the public write-up by nine weeks.
| Date | Event |
|---|---|
| May 2025 | CISA publishes advisory ICSA-25-140-03 on Danfoss AK-SM 8xxA controllers. It is updated in August 2025. |
| February 2026 | CISA publishes ICSA-26-057-10 on Copeland XWEB Pro. Copeland's fix is firmware 1.13. |
| 9 August 2026 | Claroty's Team82 publishes how both products can be taken over, and presents the work at DEF CON (Danfoss, Copeland). |
| Late August 2026 | Freezers fail at six or more US military commissaries. One base says a freezer "went into defrost mode where it heated up all the food." |
| 11 October 2026, night | First wave. Refrigeration fails at hundreds of French supermarkets. Chains named include Intermarché, Netto, Carrefour, Coopérative U, Monoprix and E.Leclerc. |
| 12 October | CERT-FR issues alert CERTFR-2026-ALE-012 naming Danfoss AK-SM and AK-SC. The Paris prosecutor opens an investigation that covers food losses and fires. |
| 13 October | CERT-FR adds Copeland XWEB and Dixell XWEB. The finance minister tells lawmakers about 700 supermarkets were affected. |
| 14 October | Censys counts 537 exposed Danfoss panels in France, down from more than 1,800 three days earlier. |
| 16 October | CERT-FR updates its recommendations and warns that campaigns are probably still running. |
| 29 October | Reuters reports the Pentagon is treating the August commissary failures as a suspected cyberattack. |
| 1 to 2 November, night | Second wave. More than 115 shops in Belgium, the Netherlands, Sweden, Luxembourg, Ireland and Estonia. Cold rooms at three sites of the French blood service EFS are also changed. No blood products are lost. |
| 3 November | Britain's NCSC says it is helping affected firms. Three meat suppliers were hit through Copeland XWEB units, and a produce wholesaler was also attacked. |
| 5 November | CISA, FBI, HHS, USDA and the Department of Transportation publish a joint fact sheet. |
| 9 November | France's foreign minister says responsibility "lies with the Islamic Revolutionary Guard Corps." |
| 13 November | VulnCheck reports 14 more flaws in Copeland and Dixell XWEB EVO, at least two already used in attacks. |
The target: one box above every fridge
A supermarket does not run each fridge on its own. Every display case and cold room has a small field controller that reads a probe and switches the compressor, fans and defrost heater. Above them sits one supervisory controller for the whole store: a Danfoss AK-SM or AK-SC system manager, or a Copeland or Dixell XWEB. It holds the setpoints, the defrost schedules and the alarm rules for everything below it, and it shows them on a web page.
That web page is the product's remote service door. The refrigeration contractor, often not the retailer, uses it to watch temperatures and tune the plant from a distance. Team82 describes the Danfoss unit as something that "effectively becomes the operational control center" of the installation. On the Copeland side the XWEB speaks Modbus over a serial line to each field controller, so whoever controls the XWEB can write to every unit on that line.
In the usual plant model this is a supervisory device with direct command of the control level beneath it. It was placed on the public internet at thousands of sites. The device that raises the temperature alarm is also the device the attacker holds.
The attack path, layer by layer.
The way in: a login that does not need the password
No authority has said which flaw was used at any victim. CERT-FR's alert names products, not vulnerabilities. What is on the public record is that each affected product family had a published route to an administrator session that needs nothing but network access to the web page.
| Product | Published route in | What it gives | Fix | State in the field |
|---|---|---|---|---|
| Danfoss AK-SM 800A | CVE-2025-41450. A hidden "code of the day" login path checks a password derived from the device's date instead of the administrator's password. | Full administrator session. Two further flaws (CVE-2025-41451, CVE-2025-41452) give command execution and control of the web server. | R4.2 for the bypass, R4.3.1 for the rest | Forescout found 80% of exposed StoreView panels below 4.3.1. |
| Danfoss AK-SM 800, AK-SC 355 | Older generation of the same family. | Danfoss says most affected French units were these models. | None. AK-SM 800 support ended in May 2021; AK-SC 355 was discontinued in 2016. | Still installed and exposed. |
| Copeland XWEB Pro (300D, 500D, 500B) | CVE-2026-25085, a login check that accepts an empty answer. CVE-2026-21718, administrator passwords derived from a fixed seed, the device's hardware address and the date. CVE-2026-24663, command execution before login. | Administrator access, then root through any of 17 further command injection flaws. | Firmware 1.13 | A note to British meat suppliers said 1.12.1 and earlier let attackers "take control without a password." |
| Copeland and Dixell XWEB EVO | 14 flaws reported by VulnCheck on 13 November. "The login page runs the submitted username as a shell command before it ever checks the password." | Command execution before login. | Copeland calls this a discontinued, end-of-life generation. | At least two flaws exploited; one of four systems hit in France. |
Two things in this table matter more than the CVE numbers.
First, two separate vendors made the same design choice: a maintenance password that is calculated, not set. On the Danfoss unit it comes from the date. On XWEB Pro it comes from a seed that is the same on every unit, plus the hardware address and the date, and the unit hands out the address and date without a login. Anyone who has the calculation has every device.
Second, the fix did not reach the field. The Danfoss update must be installed by hand. The units often belong to a contractor's estate, not the retailer's IT inventory, so the advisory may never reach the person who can schedule the work. And a large share of the installed base is out of support, where no update will ever come.
On the controller: defrost as a weapon
Once logged in, the attackers did not need an exploit for the physical effect. They used the features a refrigeration engineer uses.
- Defrost, forced or stretched. Every case has a heater that melts ice off the coil for a short period. Le Monde, cited by SecurityWeek, reported that defrost cycles were extended from one hour to as many as eight. A Belgian installer, Sabcobel, said defrost was activated remotely. A British wholesaler watched one of his fridges being put into defrost as it happened and pulled it off the internet.
- Targets raised. Setpoints were moved to 50°C. At Intermarché stores refrigeration ran at 50°C for six hours and all the food was thrown away. At the French blood service, Le Canard enchaîné reported settings of up to 110°C. That figure is the top of the range Team82 showed it could write to a Dixell field controller.
- Many changes, fast. Forescout describes "rapid successive configuration changes that suggest automation." The British wholesaler said it "was amazing how many parameters" had been altered. In the second wave settings were changed and then locked.
- The alarm arrives late. A long defrost is a legitimate state, so the controller does not treat the rising temperature as a fault until the cycle ends. The US fact sheet names the other half of this: loss of the operator's view of current temperatures. Team82 had shown in the lab that the fans can be switched off while the display keeps showing a normal reading.
- Something is left behind. CERT-FR says a factory reset "removes persistence" from a compromised unit, and that the unit stays vulnerable and will probably be taken again. The alert does not describe what that persistence is.
The result at store level was spoiled stock and, according to the Paris prosecutor, fires. Two supermarkets in Alsace put their combined loss at €270,000. There was no ransom note and no claim of responsibility. The Cloud Security Alliance's research note reads that as sabotage, not extortion.
The pace of the people probing these devices is worth noting. Decoy XWEB controllers set up by the security firm XBOW were probed within a day, and someone changed the temperature of a freezer that did not exist.
Why hundreds of stores failed in one night
A retailer plans for one store losing its cold chain. It does not plan for hundreds of stores losing it between closing time and opening time. Four conditions, all present together, turned a device flaw into a national event.
The same few products everywhere. The market for store-level refrigeration supervision is small.
Internet-facing services found on Shodan, by product family: 12,976 across 6,017 addresses in 99 countries. Source: Forescout Vedere Labs, 22 October 2026.
On public addresses. Every scanner found thousands, and France stood out in each count. Censys saw nearly 3,000 Danfoss StoreView panels on 11 October, more than 1,800 of them in France. Predicta Lab counted 2,752 the next day, 1,590 in France. Coalition's scans in the month before the attack found 54% of exposed StoreView addresses were French. These are different tools counting in different ways and do not form a trend line. The sources do not say why France had so many.
Run by a third party. Intermarché said the attack "affected a service provider" that manages refrigeration in its stores; reports of the number of Intermarché and Netto shops hit range from about 150 to about 300. The blood service's cold rooms were maintained by a Dutch rental firm, which pointed to a remote monitoring component from another supplier. In each case the door that was open belonged to someone other than the organisation that lost the stock. A contractor uses one way of connecting for all its customers, so its customers share one weakness.
At night, with the alarm in the attacker's hands. Both waves began late on a Sunday. Stores were empty, and the controller that should have raised the alarm was the device being misused. At the blood service it was an on-call technician who noticed.
The Cloud Security Alliance's research note puts the conclusion well: refrigeration at this scale "behaves more like a shared cloud dependency than like a collection of separate appliances." Loss models that treat each store as an independent risk do not describe what happened on 11 October.
What changed between the two waves
The first wave was concentrated: one country, the largest chains, one night. The response was just as concentrated. In three days the number of French Danfoss panels Censys could see fell from more than 1,800 to 537. The count outside France barely moved. Patched and unpatched units vanished at the same rate, which Censys reads as operators pulling the plug, not installing the fix.
That is the opening the second wave used. Three weeks later the same technique appeared in six countries that had watched the French attack and, by the scan data, had not acted on it.
Shops reported hit on the night of 1 to 2 November, as given by each operator; an IKEA store in Mons was also affected. Source: BleepingComputer, 2 November 2026.
Three differences from the first wave stand out.
- Smaller victims. Alongside the chains were butchers, a furniture store, British meat suppliers and a produce wholesaler. A British trade body called its members "collateral" because they ran the same software.
- Copeland in front. The British cases came through Copeland XWEB units on firmware 1.12.1 or older. Copeland's fix had been out since February and Team82's write-up since August.
- A target that is not food. The French blood service showed that the same controller sits under medical cold storage. The US fact sheet lists healthcare, chemicals and transport next to food for the same reason.
One statement from the second wave deserves more attention than it got. The Belgian installer Sabcobel said the cause was the same weakness as in France, and that a software update was not enough: physical changes to the installations were needed. If that is right, the units that were disconnected in October and not rebuilt are still waiting.
Britain's NCSC said its staff worked overnight to contact firms with exposed systems. Its technical chief said he was "surprised to see some of these things being connected to the internet still in 2026."
Attribution: a political statement, not yet a technical one
On 9 November France's foreign minister, Jean-Noël Barrot, said: "I am able to confirm that responsibility for this act lies with the Islamic Revolutionary Guard Corps." He called it "the first large-scale act of sabotage on our national territory." France summoned Iran's ambassador and said it would seek European sanctions.
That is a government attribution and should be reported as one. What it rests on has not been published. TechCrunch's account cites no evidence beyond unnamed sources. Britain has not attributed the attacks. A Belgian agency spokesperson said "it may never be possible to identify" who was responsible. US officials have reportedly linked the commissary failures to Iran, without a public statement.
There is a second candidate set that the record supports with more detail, though for earlier events. Forescout counts more than 20 claimed attacks on refrigeration systems since January 2026 by two hacktivist groups, Z-Pentest Alliance and NoName057(16), in Spain and Italy among other places, and, citing a Danfoss bulletin, describes attacks on Russian and Israeli supermarkets through the AK-SM 800 in 2022. Neither group claimed the October or November waves.
For a defender the question matters less than it seems. The technique needs a search engine for exposed devices and a published calculation. It is within reach of a state service, a hacktivist crew or an individual, and the defence is the same for all three.
What a defender would see
No authority has published indicators for these attacks. The leads below are drawn from the published research on how each login bypass works and from what victims described. They are places to look, not confirmed signatures.
At the network edge
- Any inbound connection to the controller's web or SSH port from an address that is not the refrigeration contractor's. On a correctly placed controller this count is zero.
- Requests to the Danfoss xml.cgi endpoint with the getauth action from unknown addresses, and any use of the load_nginx_headers action, which replaces part of the web server's configuration.
- On XWEB Pro, requests that read the device's hardware address or date without a session, followed by an administrator login. That is the order the password calculation needs.
- Logins to the XWEB's SSH or browser-shell service. Field engineers rarely need them.
On the controller
- Defrost schedules longer than the plant's design, or defrost running on every case at once.
- Setpoints outside any range a food or medical cold room would use.
- Dozens of parameter changes within minutes, outside working hours.
- New or changed user accounts, changed mail settings for alarms, and alarm thresholds or delays that have been widened.
- On Danfoss units, a security level that is not Strict, or default passwords still in place. CERT-FR names both.
From outside the controller
- A temperature probe that does not report through the supervisory controller and disagrees with it. This is the only check the attacker cannot quiet from the web page.
A site that finds any of the first group should assume the controller has been entered, follow CERT-FR's compromised-device steps, and compare the full configuration with a known-good copy before trusting it.
What to do, step by step of the attack
Each action below breaks one step. They are in the order a site should do them, which is CERT-FR's order.
| Attack step | What breaks it | Who has to act |
|---|---|---|
| Finds the panel with an internet scan | Take the controller off public addresses. Check the insurance terms before cutting the contractor's data feed, as CERT-FR advises. | Retailer, with the contractor |
| Logs in without the password | Firmware R4.3.1 or later on AK-SM 8xxA, 1.13 on XWEB Pro. Set Danfoss security level to Strict and change default passwords. | Contractor |
| Logs in to a unit with no fix (AK-SM 800, AK-SC 355, XWEB EVO) | Isolate now and plan replacement. A factory reset on its own leaves the unit open to the same attack. | Retailer |
| Comes back through the contractor's route | Remote access only through a VPN or protected network, with named accounts and a second factor. Filtering by the contractor's address is a stopgap, not a fix. | Contractor and retailer IT |
| Changes setpoints and defrost | Set limits on the field controllers so the supervisor cannot command a value outside a safe range. Keep a known-good copy of the configuration and compare after any incident. | Contractor |
| Delays or hides the alarm | Independent temperature monitoring that does not pass through the supervisory controller. | Retailer |
| Does it at every site at once | List every supervisory controller, including the contractor's. Ask in writing how each one is reached. Write the manual override into the store's night procedure. | Retailer |
The first and the last rows are the ones only the owner can do. The Cloud Security Alliance's note observes that these controllers often sit outside the retailer's security programme and its asset inventory, because a contractor installed and runs them.
What the sources do not say
- Which flaw was used. CERT-FR's alert names products only. Neither Danfoss nor the French authorities have tied the attack to the Team82 findings. Weak or default passwords on exposed units would explain the same result.
- How many stores. Early press counts were 250 to 350. The finance minister said about 700. Chain-level figures do not add up to either.
- What was left on the devices. CERT-FR refers to persistence without describing it.
- Whether the United States is part of the same campaign. The commissary failures look the same from the outside. No link has been confirmed.
- The evidence for Iran. Stated by a minister, not shown.
- Total losses. Individual figures exist: €270,000 for two stores in Alsace, about €15,000 for one Belgian butcher. No source gives a total.
Sources
Official
- CERT-FR, alert CERTFR-2026-ALE-012, 12 October 2026, revised 13 and 16 October
- CISA, FBI, HHS, USDA, DoT, joint fact sheet on industrial refrigeration and cold chain monitoring, 5 November 2026
Research
- Claroty Team82, Freeze the controller, defrost the food, 9 August 2026
- Claroty Team82, Chilling discoveries: Copeland XWEB Pro, 9 August 2026
- Forescout Vedere Labs, Cold chain, open control, 22 October 2026
- Forescout Vedere Labs, From supermarkets to military bases, 4 November 2026
- Cloud Security Alliance, research note on correlated refrigeration risk, 3 November 2026
- Zscaler, The fix has been out a year
Reporting
- The Register, 15 October 2026
- SecurityWeek, 15 October 2026
- Cybersecurity Dive, 13 October 2026
- BleepingComputer, 12 October 2026
- BleepingComputer, 2 November 2026
- Infosecurity Magazine, on the French blood service
- The Record, 3 November 2026
- TechCrunch, 9 November 2026
- CyberScoop, 13 November 2026