IntelliThreat — Site Header (Fable 1.0)
30-Day Trial

ICS Security Services for Controllers That Face the Internet

On the night of 11 October 2026, someone changed a setpoint.

They did it to industrial refrigeration controllers at more than 700 retail sites across France, all at once, and told them to defrost continuously. Within six hours, cabinets that should have held 4°C were at 50°C. A supermarket in Normandy reported 80°C. Equipment melted. At least 15 fires started. Food losses ran into tens of millions of euros, and the victims were not only large chains — butchers, a slaughterhouse and cooperative food-processing plants were hit too.

On 9 November, France’s Foreign Minister Jean-Noël Barrot named the attacker at a G7 meeting in Paris: the Shahid Kaveh unit, affiliated with the Electronic and Cyber Center of Iran’s Islamic Revolutionary Guard Corps. The Paris chief prosecutor issued international arrest warrants for three Iranians. France’s foreign ministry called it the first large-scale cyberattack carried out purely for sabotage on French national territory. Iran called the accusation baseless.

Strip away the groceries and look at what actually happened. An internet-reachable controller accepted a command from an unauthenticated stranger and changed the physical state of the world. That is the entire industrial control security problem in one sentence. The freezer is incidental.

Most Tools Watch the Network. The Attack Was in the Setpoint.

A controller that accepts a setpoint from the internet does not care whether it governs a freezer, a chlorine dose, a furnace or a pump. Shahid Kaveh is the same lineage previously linked to attacks on industrial controllers at US water utilities. The unit did not retool to attack supermarkets. It pointed an existing playbook at a different device class and the playbook worked, because the device class was the same: small, cheap, internet-facing controllers running firmware nobody updates.

Most security programmes are built to protect data. This attack never touched data. It changed a number, and the number moved a compressor.

The Numbers That Define the Exposure

  • 700+ retail sites compromised in a single night across France, 11–12 October 2026.
  • At least 15 fires, with food losses in the tens of millions of euros.
  • 50°C after six hours in cabinets designed to hold a few degrees above freezing; 80°C at one site.
  • 11 more countries hit in a second wave on 2 November — Belgium, Luxembourg, the Netherlands, Italy, Sweden, Estonia, Ireland, Portugal, Poland, Spain and Germany.
  • 5 federal agencies — CISA, the FBI, HHS, USDA and the Department of Transportation — issued a joint fact sheet on 5 November 2026 warning that this activity is ongoing and global.
  • CVSS 8.2 for CVE-2025-41450, the authentication bypass in Danfoss AK-SM controllers at the centre of the campaign, alongside CVE-2025-41451 (command injection) and CVE-2025-41452.

Read the last one against the first. A single authentication bypass, published and catalogued, was enough to produce physical fire damage at scale on another continent. Nothing exotic was required.

What the Attackers Actually Did

The joint advisory describes the method plainly: authentication bypass, code injection, or weak configuration on unpatched or internet-accessible controllers. No zero-day. No custom implant.

They found the devices with search engines. Controllers answering on ports 80 and 443 with recognisable management-interface fingerprints are indexed by the same public scanning services any defender can query. Honeypot data recorded authentication attempts against exposed refrigeration controllers building from 20 October and peaking on 26 October — automated scanning ran ahead of targeted exploitation, in the open, for days.

They used credentials that were never changed. The advisory’s first instruction to operators is to change all default credentials, which is instruction number one because it is failure number one. Installation manuals for this equipment are freely downloadable.

They changed the setpoint and suppressed the evidence. This is the part that matters for anyone choosing a security control. The agencies warn that attackers can force continuous heating or defrost cycles, cause inventory loss and equipment hazards, alter the service life of the system — and create a denial-of-view condition. The operator’s screen stops reflecting the plant. It is the same move that showed up in water SCADA attacks earlier this year, where modified logic disabled safety shutdowns while HMI displays kept reporting normal.

If the display is part of the attack surface, every control that reads from the display is already compromised. Alarm thresholds. Operator rounds. Shift handover. All of it.

Twenty More Vulnerabilities, Two More Vendors, One Week

The campaign is not the end of it. The device class itself is now under research pressure, and the findings are landing faster than fixes.

On 16 November, VulnCheck disclosed 14 vulnerabilities in Copeland XWEB EVO refrigeration controllers, and threat actors began probing two of them almost immediately (SecurityWeek). The same day, Midnight Blue disclosed six flaws in Danfoss AK-SM800A controllers permitting remote takeover with no login at all. No patch exists, and firmware 5.4.0 does not fix them. Cyber agencies in Belgium, Ireland and Germany told operators to act anyway (The Hacker News).

Twenty vulnerabilities, two vendors, one week, active probing within days of disclosure, and nothing to install.

That last clause is the whole argument. The standard advice assumes a patch exists and a window exists to apply it. Here, for the Danfoss flaws, neither is true today. France’s CERT-FR had already named Danfoss AK-SM, AK-SC, Copeland and Dixell XWEB controllers as confirmed targets in alert CERTFR-2026-ALE-012. The affected population is not a niche.

“Take It Off the Internet” Is Correct. It Is Not a Plan.

The joint advisory’s mitigations are sound: remove controllers from direct internet access, isolate operational technology on a dedicated VLAN, change default credentials, require multi-factor authentication, apply firmware updates, define sensible setpoint limits downstream of controllers, and alert on any deviation.

Every operator should do all of it. Then look at what remains.

The controller is still reached through an integrator’s remote-support path, because that is how it gets serviced. The firmware still has no fix for six of the flaws disclosed this week. The VLAN still carries the engineering workstation that talks to the controller, and that workstation still browses the web. The segmentation project is real, funded and eighteen months out, and the arrest warrants France issued do not slow the next campaign down.

What is left between the mitigations and the incident is detection. Specifically, detection that does not depend on the layer the attacker edits.

ICS Security Services That Assume You Cannot Patch

IntelliThreat is built for environments where the equipment cannot be touched. Self-Healing MNDR and Autonomous SIEM inspect every packet and log, 24/7, with full metadata and payload analysis across endpoints, servers, ICS/OT and IoT.

Nothing installs on the controller. There is no agent for a refrigeration controller, a PLC or an HMI, and no maintenance window to negotiate with operations. Detection sits on the network around the device.

A denial-of-view does not blind the network. An attacker who alters what a management interface reports does not alter the traffic crossing the network to reach it. The session that carried the setpoint change is observable even when the resulting display is not trustworthy.

Investigation is autonomous. Every signal is enriched and correlated to determine real risk — the Tier-1 work that takes an analyst hours, done in milliseconds. Measured speed to containment is 2.9 seconds: detected at T+0, correlated at T+1.4s, contained at T+2.9s.

Plain-language alerts. A site with no security certification on staff gets a verdict it can act on rather than a packet capture to interpret. A facilities manager is not a malware analyst and should not have to become one at 2am.

98% less alert noise in the first 30 days. A two-person team cannot work a queue. It can work a short list.

Private AI. The AI runs inside your own environment and your operational data never leaves your tenant — the first question any operator under federal or state oversight has to answer about a monitoring vendor.

ICS/OT threat hunters. IntelliThreat’s U.S.-based Threat Hunters are trained to defend ICS and OT environments, and the optional 24/7 U.S.-based SOC covers the hours a single-site facility cannot staff.

Where This Stops

Any vendor claiming network monitoring solves industrial control security is selling something. The boundary, plainly, because an operator making a safety-relevant decision is owed it.

What it sees. Traffic crossing the segments where sensors are placed: a management session from an address that has never appeared before, a configuration or firmware write, a controller reaching outbound to somewhere it should not, the scanning and enumeration that runs ahead of all of it.

What it does not see. Serial links that never touch Ethernet are invisible to a network sensor. Placement governs coverage — a sensor on the business network does not see an isolated process segment, and tap placement is an architecture decision made before deployment, not a setting afterwards. An attacker operating entirely inside a controller, generating no network traffic, is outside what any network-based approach can observe.

What it is not. This is not an OT asset inventory and discovery platform and does not replace one. It does not reverse-engineer vendor protocol semantics to judge whether a particular setpoint write is legitimate process behaviour.

The honest version is narrower than the category’s marketing and more useful than it: detection around equipment you are not permitted to patch, touch or take offline, deployed without installing anything on that equipment.

The First 30 Days

Independent of any vendor decision, in rough order of risk removed per hour spent:

  1. Inventory every controller that can be reached from the internet. Refrigeration, HVAC, building automation, process control. Include the ones an integrator connected years ago and the cellular modems in front of them.
  2. Query the public scanning services against your own address ranges. The attackers used them to build the target list. The same search tells you what they found.
  3. Change every default and shared credential on the controllers and on the equipment in front of them, and record where those credentials now live.
  4. Put remote support behind a gateway or VPN restricted to known devices, rather than exposing the controller’s management interface.
  5. Define setpoint limits downstream of the controller and alert on deviation, exactly as the joint advisory instructs. A limit enforced below the controller survives the controller being lied to.
  6. Name the thing that would tell you. If an attacker altered your displays tonight, what independently reports that the plant disagrees? If the answer is the display, that is the gap.

Steps one through five are hygiene. Step six is the one most operators cannot answer, and the one the French supermarkets answered the hard way.

Common Questions

What is ICS security? ICS security protects industrial control systems — the PLCs, HMIs, RTUs, historians and specialised controllers that run physical processes. It differs from IT security in what it optimises for: availability and safety outrank confidentiality, patch windows are rare or nonexistent, and equipment lifecycles run for decades rather than years. A compromise is measured in physical consequence, not records exposed.

What is ICS security monitoring, and how is it different from IT monitoring? ICS security monitoring watches the traffic to and from control equipment for configuration changes, unexpected management sessions and command activity that should not be happening. IT monitoring assumes you can patch, reboot and install agents on the thing you are protecting. In an industrial environment you frequently can do none of the three, which shifts the weight onto network-level observation.

What ICS security tools work on equipment that cannot be patched? Network-based detection, enforced setpoint limits downstream of the controller, segmentation, and gateway-controlled remote access. The common property is that none of them require changing software on the controller. When a vendor has no fix available — as with six of the Danfoss flaws disclosed this week — compensating controls are not a lesser option, they are the only option.

Our controllers are refrigeration, not a power plant. Is this really our problem? The joint advisory covers healthcare and public health, chemical, food and agriculture, and transportation, and the agencies specifically name retail grocery and cold storage. CISA’s guidance was blunt: every organisation that depends on refrigeration and monitoring systems should assume it could be a target. The French victims included butchers and food processors, not only national chains.

We have no security staff on site. Is autonomous detection realistic? That is the deployment assumption, not an exception to it. Connect in minutes, nothing installed on control equipment, plain-language verdicts, and an optional 24/7 U.S.-based SOC for the hours nobody is on site.

See It on Your Own Network

Start a 30-day trial and watch IntelliThreat triage your real traffic. 30 days free. No credit card required. Plans for teams of 10 to 10,000.

Or request a demo and we will walk your control network before you connect anything.

Related reading: critical infrastructure cybersecurity and why manufacturing is a target.

PDF Download

Your download is ready

 

You’ll be redirected straight to the PDF. We never share your info.

Product