tl;dr
- Overnight on 11 to 12 October 2026, refrigeration controllers at about 700 French retail sites were set to heat, destroying stock and starting at least 15 fires.
- A second wave on the night of 1 to 2 November hit more than 100 sites in at least 10 European countries, and cold rooms at three French blood-service sites.
- France attributes the October wave to Shahid Kaveh, a unit of Iran’s Revolutionary Guard. A pro-Russian hacktivist claim made on 3 November is unverified.
Two nights, 800 stores
On the night of 11 October 2026, the fridges and freezers in hundreds of French supermarkets started to warm up. By morning, cases that should hold chilled food at a few degrees were near 50°C, stock was in the bins at Carrefour, Leclerc, Intermarché and their rivals, and some stores had fires. Three weeks later it happened again, across Belgium, the Netherlands, Sweden and other European countries, and this time the cold rooms of France’s blood service were reprogrammed to 110°C before staff caught it.
The attackers took over each store’s refrigeration management. Each display case and cold room has a small controller that runs its temperature and its defrost cycle, the few minutes when the case stops cooling and a heater melts ice off the cooling coil. Above them sits a system manager, which Danfoss calls the brain of the store’s refrigeration: it logs temperatures, raises alarms and pushes settings to every case controller at once. Get into the system manager and you have every case in the store.
129 posts, and they don’t agree
Obstracts collected 129 posts about the attacks from 48 sources between 11 October and 10 November. Most are news stories repeating each other, and they disagree. Depending on which one you read, the cases reached 50°C, 55°C or 80°C, the attackers were pro-Russian hacktivists or Iran’s Revolutionary Guard, and the controllers were made by Danfoss, Copeland or Dixell.
Eight of the 129, from seven sources, are primary reports, written by someone who saw something first-hand:
| Source | What it saw | What it lacks |
|---|---|---|
| CERT-FR (9 Nov) | France’s government CERT, with the country’s attribution on behalf of five government agencies: the unit, its sponsor, its personas, and the evidence | How the attackers logged in, and any indicators |
| Sekoia (10 Nov) | A security vendor’s four honeypots posing as the targeted controllers: the reconnaissance, password spraying, and an exploit attempt on the night of the second wave | Proof that what hit the honeypots is what hit the stores |
| Pen Test Partners (14 Oct) | A UK security consultancy on how the controllers work, and how a fridge becomes a heater | Anything about who did it |
| Censys (14 Oct and 2 Nov) | An internet-scanning company’s count of every exposed controller, its firmware version, and how many vanished after each wave | Which of them were attacked |
| Claroty Team82 (13 Oct) | An industrial-security research team on the three published vulnerabilities in the controller | Evidence they were used |
| CCB (3 Nov) | Belgium’s national cyber agency on the second wave, naming Danfoss, Dixell and Copeland controllers | Attribution |
| Unit 42 (6 Nov) | Palo Alto Networks’ research team on the size of the second wave, and the hacktivist claim | First-hand evidence for either |
Even these eight don’t line up, though a few threads run between them. Four of the reports name the same Danfoss login bypass, CVE-2025-41450, and CERT-FR and Sekoia both name the same IRGC unit. But three of them describe the attack as an operation in its own right, under three different names, and two treat October and November as one event. Sekoia’s report puts both waves on Iran, when France only blamed Iran for October. CERT-FR’s covers two IRGC units, Shahid Kaveh and Shahid Shoostari, the unit France blames for its attacks on Charlie Hebdo and on a display company during the 2024 Olympics, and only the first had anything to do with the fridges. And the hacktivist claim sits in Sekoia’s and Unit 42’s reports beside Iran, with nothing to say one was assessed and the other wasn’t.
Read together, with track kept of who said what, they tell a clearer story.
Who France blames
On 9 November, CERT-FR, speaking for France’s Cyber Crisis Coordination Centre, named the IRGC as the sponsor of the October attack and Shahid Kaveh, a unit of its Electronic and Cyber Centre, as the one that carried it out. It gave four reasons:
- the attack infrastructure was also used, in the same period, to probe Schneider Electric equipment over Modbus, a common industrial control protocol, on port 502, something the unit has done since at least April 2026;
- the operators found their targets with FOFA, an internet scanning service the unit uses regularly;
- the unit’s propaganda outlet, CyberbanNews, publicised the attack from 13 October;
- traces of Farsi were found in the attack infrastructure.
Shahid Kaveh rarely uses its own name. CERT-FR lists the hacktivist personas that claim its work: CyberAveng3rs, Soldiers of Solomon, Mr. Soul and Cyber Isnaad Front. CyberAveng3rs is the name behind the November 2023 campaign against Unitronics industrial controllers, which hit a US water utility in Aliquippa.
November is a different matter. Nobody has attributed the second wave. Sekoia says the timing, infrastructure, targeting and tradecraft are close enough that the same actor may be behind both, and that is as far as any source goes.
The hacktivists who claimed it
On 3 November, the day after the second wave, two pro-Russian hacktivist groups, NoName057(16) and Z-Pentest, claimed the refrigeration attacks on Telegram, without separating the two waves. A lot of the 129 posts repeated the claim, and some still lead with it.
Unit 42 calls the claim unverified. Sekoia is blunter: NoName057(16) has a history of misleading claims, it is known for DDoS attacks rather than industrial intrusions, and the credentials it released look like aggregated infostealer logs rather than a stolen database. France named Iran six days later.
The claim is still worth recording. The next analyst who hears that NoName057(16) froze Europe’s supermarkets should find the claim, who checked it and what they found, rather than starting again. In the graph above, both groups sit beside the two waves with that assessment attached, and with no line to either wave.
How a fridge becomes a heater
France’s attribution says who. The sources agree far more on how, and they all point to Danfoss system managers, the AK-SM and AK-SC series, sold with a web interface called StoreView. Many were on the internet, Pen Test Partners explains, often through a cellular modem fitted so the refrigeration contractor could reach them, on a connection the retailer didn’t own or know about.
Put in the order the sources describe it, the October attack runs in seven steps, mostly in ATT&CK for ICS, the version of ATT&CK for industrial equipment. If Attack Flows are new to you, Beyond the ATT&CK Matrix: How to Build Dynamic Attack Flows with STIX explains how to read one.
- The operators find exposed controllers with FOFA (Scan Databases, T1596.005).
- They enumerate each controller without logging in (Remote System Information Discovery, T0888). A partner told Sekoia this began more than a week before the attack. Sekoia’s honeypots later captured the requests:
read_unitsandread_date_timeactions sent to the controller’sxml.cgiinterface, which return the units the system manager runs, sometimes the store’s name and location, and the controller’s own date. - They get administrator access to the exposed system managers (Internet Accessible Device, T0883). How is the open question of the next section.
- They force defrost on and raise the setpoints (Modify Parameter, T0836). Changing a case’s target temperature only lets it drift to room temperature. Reaching 50°C means keeping the defrost heaters on by defeating the temperature and time limits that normally end a defrost, which are ordinary settings in the Danfoss manuals.
- They change the alarm thresholds so maintenance staff aren’t paged for several hours (Alarm Suppression, T0878).
- Stock is destroyed at about 700 sites (Loss of Productivity and Revenue, T0828).
- At least 15 fires start (Damage to Property, T0879). Pen Test Partners explains them as a side effect: heaters built to run for minutes ran for hours, and overheated where there was already a weak connection.
The last two steps are worth keeping apart. A retailer can recover from spoiled stock, but a fire turns a cyber incident into a safety one.
The way in nobody has confirmed
Every source agrees the system managers were reachable from the internet, and none confirms how the attackers logged in.
The leading suspect is CVE-2025-41450, disclosed in 2025. Older Danfoss firmware accepts a second, undocumented login: send the right code to xml.cgi and you get an administrator session without a valid account. The code is derived from the controller’s date by a fixed algorithm, so it is the same on every unit on a given day, and the controller will tell anyone its date without a login, which is exactly what read_date_time asks for. Danfoss removed the code in firmware 4.2. Version 4.3.1 fixes two later bugs that let a logged-in user run commands or disrupt the interface, CVE-2025-41451 and CVE-2025-41452.
Plenty of controllers were still on the old firmware. Censys counted 1,825 StoreView panels online in France on 11 October, and 1,158 of them ran firmware older than 4.2. By 14 October there were 537, and by 2 November, 220. Every firmware group shrank by about 70% in the first three days, so operators appear to have unplugged their controllers rather than patched them.
Sekoia couldn’t confirm the date code was used either. Its honeypots did see read_date_time requests, which an attacker would send first, and password spraying with 12345 against the Supervisor and admin accounts. On the night of the second wave, two addresses tried something else: an unauthenticated request to rename user ID 1, the full administrator. Sekoia suspects this is the first half of a non-public exploit chain, renaming the account to a known login so a second request can reset its password. The CCB adds unchanged default passwords to the list.
So the three vulnerabilities stay suspects. The graph shows them beside the access step, with what each source found, and nothing yet connecting them to it.
What defenders can hunt for
The attackers left very little to block. CERT-FR published no indicators, and the only addresses published in any primary source are the two that tried the rename exploit on Sekoia’s honeypots: 178[.]83[.]226[.]215 and 37[.]16[.]75[.]128. They were seen on a honeypot, not on a victim, and Sekoia only says they may be linked to the second wave, so treat them as leads rather than proof. Both are in the reference section at the end.
The behaviour is more useful than the addresses. The enumeration and the password spraying Sekoia captured all went to one endpoint, xml.cgi, and nobody outside the store, its contractor or Danfoss’s cloud service has a reason to call it. A Sigma rule is a YAML file that describes suspicious activity in a log, which detection tools then convert into queries for a specific SIEM. If you are new to Sigma, Writing Effective Sigma Detection Rules: A Guide for Novice Detection Engineers explains each field. This one is for the web or reverse-proxy logs in front of a system manager:
title: External Request to a Danfoss System Manager XML Interface
id: 80ed4d02-ca1d-5c04-bfc9-7da0873ce7b7
status: experimental
description: Detects requests to the xml.cgi interface of a Danfoss AK-SM or AK-SC refrigeration system manager from outside the networks that should manage it. In the October and November 2026 attacks on European supermarkets, unauthenticated read_units and read_date_time requests to xml.cgi were used to enumerate controllers, and password spraying against the Supervisor and admin accounts went to the same endpoint.
references:
- https://www.cert.cyber.gouv.fr/cti/CERTFR-2026-CTI-009/
- https://www.sekoia.com/blog/the-cold-chain-on-fire-unpatched-danfoss-controllers-exploited
- https://www.pentestpartners.com/security-blog/you-dont-need-a-nation-state-to-warm-up-a-fridge/
author: dogesec
date: 2026-11-10
tags:
- attack.discovery
- attack.t0888
logsource:
category: webserver
detection:
selection:
cs-uri-stem|endswith: '/xml.cgi'
filter_maintenance:
c-ip|cidr:
- '10.0.0.0/8'
- '172.16.0.0/12'
- '192.168.0.0/16'
condition: selection and not filter_maintenance
falsepositives:
- The refrigeration contractor's supervisory software or Danfoss cloud connector polling the system manager from a fixed address; add those addresses to the filter
- Internet scanners such as Censys and Shodan indexing exposed controllers, which still means the controller is exposed
level: high
The rule can’t see the action inside the request body, so it fires on any outside request to the endpoint. Any hit means the controller is reachable from somewhere it shouldn’t be, which was the root cause of both waves. The filter excludes private address ranges; add the contractor’s fixed addresses to it.
The rule catches step 2, the enumeration. The password spraying and the rename attempt went to the same endpoint, but only on Sekoia’s honeypots, so it may catch the break-in too. Nothing catches the step that heated the fridges: steps 4 and 5 happen inside the controller’s settings, and the alarms that would have shown them were the thing the attackers changed.
Both waves in one graph
Put together, this is the whole case: the eight reports, the two waves, the unit France blames and its sponsor, the claim nobody has verified, the seven steps, the vulnerabilities nobody has confirmed, the honeypot addresses and the rule.
A hunter gets one thing to look for: outside requests to xml.cgi. An analyst gets the state of the case: France blames Iran for October, nobody has been blamed for November, the hacktivists’ claim is unverified, and the way in is still unknown. Each of those sits in the graph with its source.
In summary
The fridge attacks produced 129 posts in a month, and most of them repeated each other, disagreed on details or passed on a claim that didn’t hold up. Eight primary reports each saw one part: France saw who, Sekoia’s honeypots saw how the attackers probed, Censys saw what was exposed, and Pen Test Partners and Claroty explained the controller. Read together, they give one picture that keeps France’s attribution to October, keeps the hacktivist claim and the suspected vulnerabilities as open questions, and ends in a Sigma rule that covers discovery, with nothing able to see the step that heated the fridges.
Reference: IoCs and ATT&CK techniques
IoCs
| Type | Value | What it is | Linked to | Confidence | STIX ID |
|---|---|---|---|---|---|
| IPv4 | 178[.]83[.]226[.]215 |
Sent an unauthenticated user-rename exploit request to Sekoia’s Danfoss honeypots on 1 Nov 2026; not confirmed for either wave | Possibly the second wave (Sekoia) | Low | indicator--e512e86c-8d52-570f-855f-b47230edc359 |
| IPv4 | 37[.]16[.]75[.]128 |
Sent an unauthenticated user-rename exploit request to Sekoia’s Danfoss honeypots on 1 Nov 2026; not confirmed for either wave | Possibly the second wave (Sekoia) | Low | indicator--4d8b539e-d9c8-5ac3-a525-409030750ef2 |
Techniques (October wave Attack Flow)
| Step | Action | Technique | Tactic | STIX ID |
|---|---|---|---|---|
| 1 | Exposed controllers found with FOFA | T1596.005 | TA0043 | attack-action--93deb330-a463-5260-b9e3-0cecb02308d7 |
| 2 | Unauthenticated controller enumeration | T0888 | TA0102 | attack-action--0467d543-a2ce-5ded-99cc-7cfd553500f4 |
| 3 | Administrator access to internet-exposed system managers | T0883 | TA0108 | attack-action--faa7e8df-38e3-5c6c-b377-cd58541c3625 |
| 4 | Defrost forced on and setpoints raised | T0836 | TA0106 | attack-action--4466d0ee-efde-52ec-9c20-b685ef0a38ea |
| 5 | Alarm thresholds changed so nobody is paged | T0878 | TA0107 | attack-action--ac82f160-e31d-5676-b163-b84227a64908 |
| 6 | Stock destroyed at about 700 sites | T0828 | TA0105 | attack-action--426f172f-a96a-5b31-966f-de956c5dec9a |
| 7 | At least 15 fires | T0879 | TA0105 | attack-action--ca199204-c9d3-5831-834c-7f298ffb8ea8 |
The Sigma rule is indicator--80ed4d02-ca1d-5c04-bfc9-7da0873ce7b7, and the October campaign is campaign--23ba9688-5561-5c12-aa78-4bcc38da6cda.
Get new posts by email
New threat research and tutorials from dogesec, plus the weekly Obstracts threat brief. No spam, unsubscribe any time.
