Europe's fridge attacks, up close: 110 °C typed into Copeland XWEB from an iPhone
Since October 2026, refrigeration controllers in supermarkets across Europe have been taken over and their settings changed, first in France and then in other countries15. On the evening of 13 November, the same kind of takeover played out on our own decoys: someone using Chrome on an iPhone logged in to two of our Copeland XWEB refrigeration decoys with the factory password. Over the next hour they typed the same block of settings into one controller after another: set point 110 °C, every alarm limit at the top of its range, defrost stretched as far as it goes. Then they switched off the lights, muted the alarms and locked the keypads. This is that session, value by value, and what to check on your own units.
In brief
- The login was Admin / Admin, the factory default, accepted on the first try on both units. Nothing was exploited.
- In about an hour the visitor made 39 parameter writes and 103 commands across the cold rooms, cases, freezer islands and compressor racks the two units supervise.
- The block pushes every field to the highest value it accepts: set point 110 °C, alarm thresholds 255, alarm delay 255 minutes, defrost up to 50 °C for up to 255 minutes.
- Then, controller by controller, three commands: Light OFF, Alarm Mute, KeyboardLock.
- It was typed by hand on a phone. Six of the writes carry
11”where 110 was meant. - The same login session was presented from five addresses: four at hosting providers and one at an Iraqi internet provider.
- Two days later the same phone opened a third decoy and typed in two XWEB EVO flaws that had been public for three days: the username command injection (CVE-2026-124002) and the hard-coded Administrator password (CVE-2026-124004).
The first unit: 36 minutes
An XWEB is the supervisor that sits above a store's refrigeration: one web interface for every case, cold room and compressor rack on the site. Our decoys present twelve Dixell controllers each. The interface pages are our own build of the product's front end, so the requests below were written by the page; what the visitor chose was the password, the values and the order. All times are UTC.
| Time, 13 Nov | What the visitor did |
|---|---|
| 21:18:42 | Opened the login page. |
| 21:18:49 | Logged in as Admin / Admin. Accepted on the first try. |
| 21:18:57 to 21:20:47 | Opened the parameter pages of two controllers and the user list, four times. |
| 21:22:35 to 21:23:37 | Controller 1, a first pass: Light OFF, Alarm Mute, then the first block of settings (below), KeyboardUnlock, Alarm Mute again. |
| 21:30:59 to 21:43:21 | The same block into controllers 2 to 12, one controller every minute to minute and a half. |
| 21:44:24 to 21:48:04 | Fan mode (FnC) set to 3 on ten controllers. |
| 21:49:25 | Back to controller 1: set point and minimum set point to 150, above the 110 the field accepts. |
| 21:51:41 to 21:51:59 | User list again, then the alarm list. |
| 21:52:18 to 21:54:54 | A sweep of the whole site in two and a half minutes: Alarm Mute and KeyboardLock on all twelve controllers, Light OFF on eleven. |
The block
This is what was typed into controller 1, and then, with small variations, into every other controller:
SEt 110 Hy 25 LS 110 US 150 ot 12
dtE 50 dtS 50
ALU 255 ALL 255 ALd 255 dAO 23
Later controllers got a longer version. This is controller 4, at 21:34:01:
SEt 110 Hy 25 LS 110 US 150 ot 12
odS 255 AC 50 CCt 24
dtE 50 MdF 255
ALU 255 ALL 255 AFH 25 ALd 255 dAO 23
The codes are the Dixell XR-series parameters these controllers use. Read against the ranges the interface shows, every value is the maximum its field accepts:
| Code | Parameter | Typed | What that does |
|---|---|---|---|
SEt | Set point | 110 °C | The thermostat only calls for cooling above the set point. At 110 °C it never does. |
LS, US | Minimum and maximum set point | 110, 150 °C | Nobody can bring the set point back down without first finding and changing the limit. |
Hy, ot | Differential, probe calibration | 25, 12 °C | The widest switching band and the largest probe offset. |
ALU, ALL | High and low temperature alarm | 255 | No temperature the case can reach raises an alarm. |
ALd, dAO | Alarm delay, alarm delay at start-up | 255 min, 23 h | Any alarm that did trigger would wait hours before showing. |
dtE, dtS, MdF | Defrost termination temperatures, maximum defrost length | 50 °C, 255 min | A defrost runs until the evaporator reaches 50 °C, for up to four and a quarter hours. |
odS, AC, CCt | Output delay at start-up, anti-short-cycle delay, continuous cycle | 255 min, 50 min, 24 h | After a restart the outputs stay off for hours, and the compressor waits 50 minutes between starts. |
FnC | Fan operating mode | 3 | The last mode in the list. |
The three commands finish the job. With the light off, the alarm muted and the keypad locked, the case looks switched off, makes no sound, and cannot be put right from its own front panel.
Reports from the October incidents in France described the same shape of damage: refrigeration units at one chain ran at 50 °C for six hours, and defrost cycles were lengthened, which delayed the alarms1.
The second unit: a quicker routine
At 21:59:04 the same phone opened a second decoy and logged in with the same pair. By now there was a routine. For each controller: one write with the whole block, fan mode now usually included, then Light OFF, Alarm Mute and KeyboardLock within a few seconds. Eleven controllers were done between 22:03:24 and 22:15:19.
Then the visitor went back to check their own work:
| Time, 13 Nov | Controller | Follow-up |
|---|---|---|
| 22:17:15 | 1 | dtE set to 50. |
| 22:17:48 to 22:18:01 | 3 | KeyboardUnlock, LS retyped as 110, KeyboardLock. |
| 22:19:55 to 22:20:07 | 7 | KeyboardUnlock, dtE and dtS set to 50, KeyboardLock. |
| 22:20:27 to 22:20:55 | 9 | KeyboardUnlock, SEt, LS, US and dtS retyped, then Light OFF, Alarm Mute, KeyboardLock. |
The last action that evening was at 22:23:16: the alarm log of the second unit.
Typed by hand, on a phone
Nothing in the session looks scripted.
- The browser. Every request carried the same user agent: Chrome 157 on an iPhone running iOS 26.6.2. Page loads listed English, Arabic and Spanish as languages; the interface's background requests carried
aralone. - The typing mistakes. Six writes contain
11”or15”in place of 110 or 150. The last character is a typographic closing quote, the kind a phone keyboard inserts. Two controllers got an alarm threshold of 155 where the others got 255. Several fields were sent empty. - The pace. On the first unit each controller took about a minute to a minute and a half: long enough to open the page and fill in a dozen fields with a thumb.
- The corrections. Controllers 3 and 9 on the second unit, both with mistyped values, are the ones the visitor unlocked, retyped and locked again.
Five addresses, one session
Each login returned a session token, and the open pages kept refreshing the device list with it every 15 to 20 seconds. The two tokens issued that evening were then presented from five different addresses:
| Time (UTC) | Address | Network |
|---|---|---|
| 13 Nov 21:18 to 23:35 | 169.150.218[.]23 | AS212238 Datacamp, Netherlands (hosting) |
| 23:38 to 23:40 | 169.224.124[.]75 | AS199739 Earthlink, Iraq (internet provider) |
| 23:40 to 23:41 | 84.252.113[.]54 | AS212238 Datacamp, France (hosting) |
| 23:42 | 169.224.124[.]75 | Earthlink, Iraq |
| 23:42 to 23:50 | 84.252.113[.]57 | AS212238 Datacamp, France (hosting) |
| 23:51 to 23:54 | 169.224.124[.]75 | Earthlink, Iraq |
| 23:54 to 23:57 | 80.79.6[.]92 | AS49981 WorldStream, Netherlands (hosting) |
| 23:59 to 14 Nov 00:31 | 169.224.124[.]75 | Earthlink, Iraq |
Every write and every command came from the first address. After that the pages were only refreshing. The pattern is the one a VPN makes when it drops and reconnects through a new exit: each time a hosting address went away, the same pages carried on from the Earthlink address, and they stayed there for the last 32 minutes.
They came back, twice
14 November, 12:00. From 80.79.6[.]92, the phone logged in to both units again with Admin / Admin and watched the device lists for about a quarter of an hour. It changed nothing.
15 November, 22:37. From a new address, 185.107.80[.]116 (AS43350 NForce, Netherlands), the same phone opened a third decoy, logged in with Admin / Admin and looked at the user list. Six minutes later it went back to the login page and typed this:
22:43:52 username admin; sleep 5 # password jakakak
22:43:59 username Admin; sleep 5 # password jakakak
22:44:04 username Admin; sleep 5 password jakakak
22:44:29 username Administrator password xwebevonewgeneration
22:44:46 username Admin password Admin
The first three are the XWEB EVO username command injection, CVE-2026-124002: shell characters in the username field run as root before the password is checked, and a five-second pause is the test that it worked2. The fourth is CVE-2026-124004, a hard-coded password for the built-in Administrator account3. Both records were published on 12 November. Three days later a person was trying them by hand, on a phone, against a controller they already had the password for. They then logged back in the ordinary way and watched the device list until 23:23.
Scripts got there first
The hand-typed attempts were not the first use of those records. The first scripted request for the same set of flaws reached an XWEB decoy at 21:44 UTC on 12 November, under two hours after the records went public:
| First seen (UTC) | Source | Request | Flaw |
|---|---|---|---|
| 12 Nov 21:44 | 202.61.138[.]117 | POST /cgi-bin/sysinfo.lua, then POST /cgi-bin/config.lua with action=get, then GET /cgi-bin/xwconsole.cgi | CVE-2026-124011, CVE-2026-124014 and CVE-2026-124010, checked in one pass |
| 13 Nov 01:14 | 154.9.235[.]100 | GET /cgi-bin/xwconsole.cgi?sh=id | CVE-2026-124010, command execution as root4 |
| 13 Nov 07:13 | A paced scanner rotating through many addresses | GET /xwconsole.cgi?sh=;id and username=user$(sleep 8) posted to /login.lua, /cleanout.lua and /api.lua | CVE-2026-124010 and CVE-2026-124002 |
| 14 Nov 17:20 | 177.4.12[.]11 | GET /cgi-bin/xwconsole.cgi?sh=echo+V9QUN4DTV | CVE-2026-124010, with a marker to read back |
What to check on your XWEB
- Change the passwords. Admin / Admin opened both units on the first try. Change every account, and remove the ones nobody uses.
- Compare the parameters with your commissioning sheet. On each controller look at
SEt,LSandUS, thenALU,ALL,ALdanddAO, thendtEandMdF. A value sitting at the top of its range (110, 150, 255, 50, 23) is what this session left behind. - Walk the cases. A dark case with a locked keypad and no alarm sounding is the visible result of the three commands.
- Read the web logs. Look for logins from addresses that are not your maintenance provider's, for user names containing
;,$(or a backtick, for a login as Administrator you did not make, and for any request toxwconsole.cgi. - Take the web interface off the internet. Reach it through a VPN or a protected management network.
- XWEB EVO is out of support. The CVE records state the product line reached end of service in March 20263. The hard-coded Administrator password is in versions before 5.6.2 and the username injection in versions before 5.7.0, so update what you can and plan the replacement.
Indicators
| Indicator | Role | Confidence |
|---|---|---|
169.150.218[.]23 | Admin / Admin login to two XWEB decoys, 39 parameter writes and 103 commands, 13 Nov 2026 21:18 to 23:35 UTC | High for the event |
169.224.124[.]75 | Same session tokens, device list refreshes only, 13 Nov 23:38 to 14 Nov 00:31 UTC | High for the event |
84.252.113[.]54, 84.252.113[.]57 | Same session tokens, device list refreshes only, 13 Nov 23:40 to 23:50 UTC | High for the event |
80.79.6[.]92 | Same session tokens 13 Nov 23:54; new Admin / Admin logins to both decoys 14 Nov 12:00 UTC | High for the event |
185.107.80[.]116 | Same phone on a third decoy: Admin / Admin, then hand-typed CVE-2026-124002 and CVE-2026-124004, 15 Nov 22:37 to 23:23 UTC | High for the event |
Mozilla/5.0 (iPhone; CPU iPhone OS 26_6_2 like Mac OS X) AppleWebKit/605.1.15 (KHTML, like Gecko) CriOS/157.0.8103.37 Mobile/15E148 Safari/604.1 | User agent of every request in the sessions above | Context only |
User name admin; sleep 5 # with password jakakak | Hand-typed test for CVE-2026-124002 | High for the event |
Login as Administrator with xwebevonewgeneration | CVE-2026-124004, the hard-coded password | Pattern |
202.61.138[.]117, 154.9.235[.]100, 177.4.12[.]11 | Scripted requests for config.lua, sysinfo.lua and xwconsole.cgi, 12 to 16 Nov 2026 | High for the event |
xwconsole.cgi?sh= | CVE-2026-124010; worth an alert from any address | Pattern |
References
- SecurityWeek, Supermarket Refrigeration Attack Widens to Copeland, Dixell Controllers, 15 October 2026.
- CVE Program, CVE-2026-124002: XWEB EVO pre-authentication OS command injection through the user name (published 12 November 2026).
- CVE Program, CVE-2026-124004: XWEB EVO hard-coded Administrator password (published 12 November 2026).
- CVE Program, CVE-2026-124010, CVE-2026-124011 and CVE-2026-124014:
xwconsole.cgi,sysinfo.luaandconfig.lua. - Facilities Dive, European grocery stores hit with second refrigeration system hack, November 2026.