Our weekly review covers 10 to 15 October. The approach does not change: the events that matter, read through the eyes of a risk owner rather than an analyst, and for each one, what it changes for a Canadian organization.
The last issue came down to one line: your trusted tools became the way in. This week takes it one step further: the equipment nobody managed. End-of-life refrigeration controllers, connected to the internet for remote maintenance, let attackers turn up the heat in hundreds of stores. Backup servers run by service providers, a VPN appliance that was fully patched, AI tools set up by a team and then forgotten. Each time, the equipment was reachable from outside, and nobody inside had put it on the risk register under their name.
Part 1: refrigeration in 700 supermarkets, controlled from the internet
What happened
On the night of 11 October, attackers took remote control of the controllers that run refrigeration in French supermarkets. They stretched defrost cycles from about one hour to as many as eight, according to Le Monde, which delayed the alarms. At Intermarché, refrigerated units ran at 50°C for six hours and all the food was thrown out. The Paris prosecutor's office also reports fires.
On 13 October, Finance Minister Roland Lescure told lawmakers that about 700 of France's roughly 10,000 supermarkets were affected, up from 300 to 350 in the first counts. Intermarché reports about 150 stores hit, Carrefour 80, Coopérative U 80 (two of its Alsace stores put their losses at €270,000), and Monoprix and E.Leclerc are also affected. No claim, no ransom demand. Status: confirmed by France's cybersecurity agency ANSSI, the retailers and the prosecutor.
The mechanism: end-of-life equipment, exposed, maintained by someone else
On 12 October, ANSSI's CERT-FR issued an alert on Danfoss AK-SM and AK-SC controllers, extended the next day to Copeland and Dixell XWEB controllers. It says three things a leadership team should read:
- Exposing a management interface to the internet goes against good practice. The targeted devices were reachable from the internet, often on a public IP address.
- A factory reset is not enough. It removes the attacker's persistence, but an end-of-life device with no patches that stays exposed will be reinfected. The campaigns are probably still running.
- Before cutting the link to your refrigeration contractor, check your insurance contract. Remote maintenance is often a condition of coverage.
Danfoss says most affected devices are end-of-life models: support for the AK-SM 800 ended in May 2021, and the AK-SC 355 was discontinued in 2016. A few current controllers running outdated firmware were also hit.
These controllers are not run by the retailers themselves. Coopérative U's stores, for example, do not buy from Danfoss: they are customers of refrigeration contractors who supply and maintain the system, remotely. E.Leclerc said the attack hit one of its refrigeration equipment suppliers.
The most telling number comes from Censys. On 11 October, the search engine for exposed devices counted nearly 3,000 Danfoss control panels reachable from the internet, more than 1,800 of them in France. By 14 October, 537 remained in France; elsewhere, the count had barely moved. Censys notes that devices were mostly unplugged, not patched.
What it changes: in three days, more than 1,200 French devices disappeared from the internet. So they did not need to be there, or not like that. They were there because nobody had decided otherwise. A refrigeration, heating or door-access controller belongs to facilities, not IT; it is maintained by a contractor who needs to reach it; and it lasts fifteen years, long after its support ends. So it shows up neither in the IT inventory, nor in the risk register, nor in the contractor's questionnaire. The day it fails, it is not an IT incident: it is lost stock, a closed store, a fire risk. The question for leadership is not "are our fridges on the internet?" but "who, here, is entitled to answer that question?".
Part 2: the edge, again, and a record month of patches
Ivanti Connect Secure: a fully patched appliance, and yet
On 13 October, Volexity described an actor it calls UTA0576 exploiting a chain of three previously unknown flaws in the Ivanti Connect Secure VPN since early October. The compromised appliance was running version 22.8R2.4, fully patched. The attacker left no web shell and no malware: it re-ran the exploit for each command, tampered with the logs and evaded Ivanti's own integrity checker. Its targets: Active Directory credentials, LAPS passwords and BitLocker recovery keys. Two flaws are patched; the privilege escalation flaw is not yet, with a fix promised "in the coming weeks". CISA added the authentication bypass (CVE-2026-104935) to the KEV catalog on 13 October with three days to act and, if in doubt, recommends a factory reset followed by rotating every password, certificate and API key. Shadowserver counts more than 5,500 instances exposed to the internet. Status: confirmed by Ivanti ("limited exploitation"), Volexity and CISA.
An email gateway, firewalls, backup servers
On 14 October, CISA added four exploited flaws to the catalog:
- Cisco Secure Email Gateway (CVE-2026-76519, CVSS 9.8): a crafted attachment lets an unauthenticated attacker write files and run commands as root, when AMP File Reputation is enabled. It is the second exploited flaw on this gateway since mid-September. Cisco warns that an attacker with those rights can hide their tracks and advises checking external firewall logs.
- Palo Alto PAN-OS (CVE-2026-0322, CVSS 8.6): code execution through the firewalls' management interface, for an attacker who is already authenticated. Palo Alto reports "very limited exploitation" and offers no workaround other than restricting the management interface to trusted internal addresses.
- AhsayCBS (CVE-2026-105133 and CVE-2026-105134), a backup platform used in particular by service providers: authentication bypass then code execution, exploited to drop web shells and cryptominers, according to Huntress. On 12 October, Ahsay acknowledged its first patches were incomplete.
On 15 October, an unauthenticated file read flaw in self-hosted Atlassian products (Jira, Confluence, Bitbucket and others, CVE-2026-21589, CVSS 9.3) joined the list. Status: confirmed by the vendors and CISA.
More than 1,400 Microsoft fixes in one month
The 13 October Patch Tuesday fixed more than 1,400 vulnerabilities (counts range from 1,407 to 1,413 depending on the source), over 1,100 of them in Windows alone. According to SecurityWeek, this single month exceeds all of 2025, and the outlet links the surge to AI-assisted bug discovery. Only one flaw is exploited: CVE-2026-96951 (CVSS 8.8), a Mark-of-the-Web bypass that suppresses SmartScreen warnings and Office Protected View; it requires the user to open a file. Microsoft is also introducing "risk cohorts" to help prioritize.
What it changes: three of the four products exploited this week are security or edge devices, and the only protection Palo Alto offers is the one CERT-FR gives for the fridges: do not expose the management interface. Ivanti shows that a fully patched device is not enough; Ahsay, that a patch can be incomplete. The decision that holds in both cases sits upstream: what is allowed to be reachable from the internet, and who checks it every month? And with more than 1,400 fixes in one month, volume can no longer be managed by severity score: you manage it by starting with what is exploited and exposed in your environment.
Part 3: Canada, from hospital doors to tax slips
On 11 October, the INC Ransom group claimed Health Sciences Centre Winnipeg, Manitoba's largest hospital, saying it holds patient records, staff data and financial data, and disputing the hospital's statement that no sensitive data was affected. What is confirmed: in August, Shared Health reported a ransomware incident that hit facility systems, including ventilation and door controls, without being able to say at the time whether data had been taken. Status: incident confirmed in August, data theft claimed; nothing yet confirms that the October claim concerns the same incident.
On 12 October, DragonForce claimed Canadian Mennonite University in Winnipeg. The university had itself confirmed the intrusion on 2 October: detected on 18 September, it exposed T4 slips from 2015 to 2025, T2202, T4A and TD1 forms and records of employment, with social insurance numbers. The university offers two years of credit monitoring and has notified police and the Office of the Privacy Commissioner. Status: confirmed by the university, claimed by the group.
On 14 October, Akira claimed Shediac Lobster Shop in New Brunswick, announcing the release of about 10 GB of documents (employee passports and social insurance numbers, client files, contracts). In Quebec, just before the week, the Storm group claimed CETAM, an ambulance services cooperative in Montérégie, on 8 October. Status: claimed, with no confirmation from the organizations named.
Finally, on 15 October, the Burnaby School District in British Columbia confirmed that the 21 September ransomware attack involved copying personal information: names, contact details, student numbers and birth dates of students and parents, citizenship and residency documents, kindergarten assessments, lists of students receiving special support, sometimes with a diagnosis, and staff records. The RCMP and the province's privacy commissioner have been notified. Status: confirmed by the district.
The context: by our count of claims published on ransomware.live, 264 Canadian organizations have been claimed by ransomware groups since 1 January 2026, with manufacturing first at 49 claims (18.6%), ahead of professional services and technology. These are claims, not confirmed incidents. The Canadian Centre for Cyber Security, in its 2025-2026 assessment, describes "opportunistic" actors: they hit what is open, not a sector picked in advance. On 13 October, the Centre also co-sealed with nine partners an advisory on PRC state-linked actors that combine automated tools with hands-on work, whose first recommendation is to disable unused services.
What it changes: a hospital's ventilation and doors in August, ten years of tax slips at a university, lists of students with diagnoses at a school district. Both lessons meet the supermarkets'. First, facility systems are inside an attack's blast radius, just like servers. Second, what costs the most after a breach is the data you kept with no reason to keep it: every year of archive you hold is one more year to notify. Cyber Security Awareness Month devotes its third week precisely to protecting your information.
Part 4: AI on both sides of the table
On the platform side, Censys published on 15 October a count of AI software exposed to the internet: 267,915 IP addresses, up 63% in nine months. The automation tool n8n accounts for 113,926 of them, and 76.1% of those that show their version run one affected by a flaw rated 9 or higher. The same day, researchers reported exploitation of a critical flaw in Langflow, a platform for building agents (CVE-2026-105740, CVSS 9.9): more than 1,700 attempts caught by their honeypots, with secrets stolen along the way. Status: reported by researchers.
On the agent side, India's CERT-In issued an advisory on 15 October about AI agents under test by AI companies that reached live systems of Indian organizations through exposed credentials and overly permissive cloud roles, and read personal and confidential data there. None of the organizations detected it. Status: confirmed by the authority. On 10 October, Anthropic for its part cut live internet access for all its internal evaluations after models under test took unintended actions against real websites, including exploiting a command injection on a university server. Status: confirmed by the company, which describes the real-world impact as "minimal".
On the defender side, XBOW disclosed on 13 October that its AI agents had helped build a full Chrome exploit chain in six days, sandbox escape included, for about US$40,000 of compute; Google confirmed and fixed the flaws and paid a US$250,000 bounty. Humans picked the targets and assembled the chain. Update Chrome (version 152.0.7977.82 or later).
What it changes: the exposure nobody owns now has an AI version. An automation tool set up by a team to save time, connected to email and the CRM, reachable from the internet and never updated, is the refrigeration controller of 2026. And an agent given an overly broad cloud role reads whatever it can read, without anyone noticing. That is the subject of this week's AI series: an AI agent is a privileged account, so treat it like one.
The three questions of the week for an executive committee
- Which internet-connected equipment belongs to facilities rather than IT (refrigeration, ventilation, access control, production), and who, by name, answers for it?
- For each contractor that maintains equipment remotely, which path do they connect through, who can cut that link without waiting, and does your insurance contract allow it?
- How many of your exposed devices are end of life, and how fast could you take them off the internet if the news forced you to?
If these questions have no documented answer, our cyber risk score tells you where you stand in ten minutes, free and with no signup: 9 domains, a score and prioritized actions. To know what is actually reachable from the internet, on the web as well as in OT, and to track the threats aimed at your sector, that is the job of Forta Exposure and Forta Radar.
One last word, then we stop talking about ourselves
Tomorrow, our Saturday article extends this week with a twenty-minute test: the most dangerous subdomain is the one nobody deleted. On Sunday, the risk series asks who accepts, treats, watches and reports on each risk: "IT" is not a risk owner. And this recap also arrives by email, every Friday: sign up here, unsubscribe in one click.
Sources: CERT-FR, alert CERTFR-2026-ALE-012 · SecurityWeek, the supermarket attack widens to Copeland and Dixell · The Hacker News, French supermarkets · Reuters via NBC News · CISA, Ivanti Connect Secure · Volexity, the Ivanti chain · BleepingComputer, Ivanti · CISA, 13 October alert · CISA, 14 October alert · CISA, 15 October alert · Palo Alto Networks, CVE-2026-0322 · SecurityWeek, Cisco Secure Email Gateway · Help Net Security, PAN-OS, Cisco and AhsayCBS · The Hacker News, Atlassian · SecurityWeek, October Patch Tuesday · Tenable, October Patch Tuesday · ransomware.live, Health Sciences Centre · Shared Health, August incident · Canadian Mennonite University, incident notice · CP24, Burnaby · ransomware.live, victims in Canada · Canadian Centre for Cyber Security, 13 October joint advisory · Canadian Centre for Cyber Security, National Cyber Threat Assessment 2025-2026 · Help Net Security, exposed AI infrastructure · The Hacker News, Langflow · The Hacker News, the CERT-In advisory · Anthropic, unintended model actions · The Register, XBOW and Chrome