In October, hackers heated supermarket fridges in France. In November they did it again to shops in seven European countries, including a fruit and vegetable wholesaler in London. The attackers did not need AI, because the controllers that run the fridges were old, reachable from the internet and protected by logins that were easy to get. The useful question for a business owner is what AI does to the odds that something similar happens to them.
Attackers collect whatever answers
The people behind these attacks do not pick a shop and study it. They scan the internet for a certain kind of device and log into every one that answers. It works like a thief who tries every car door in a car park and drives away the ones that open.
Before the French attacks, more than 1,700 copies of the affected Danfoss software were open to the internet in France alone. In the Netherlands, the victims were mostly butchers and small grocers. A shop like that has no security team, and probably does not know the controller is online at all. The connection is there so the maintenance company can change settings from its office.
Alex Pearce, who runs the wholesaler, saw three units show 150C on their panels and pulled the network cable. He has not plugged it back in.
Old gear gets found but not fixed
AI has not invented new ways to attack. It has made it far cheaper to find weaknesses that were tolerated for years. We made the same point when we wrote that old weak spots were only safe because nobody looked.
The next step is about who gets the fix. When AI finds a flaw in a product the maker still supports, a patch follows, and the Linux kernel project alone patched more than a thousand flaws in a single month. But the Danfoss controllers that were hacked were mostly ones the company no longer updates. For that kind of device, a flaw becomes known and no fix ever arrives. Even a factory reset does not help, because a reset device that stays online will very likely be broken into again.
So the more AI is pointed at old software, the more the danger concentrates in equipment that is out of support. That list is longer than most owners think: cold-room controllers, camera recorders, door-entry panels, alarm boxes, printers and routers that a contractor installed and nobody has opened since.
Attacks have not yet risen as fast as the flaws being found, but I expect that gap to close for this category, because the work of using an old flaw is mostly the work of finding which devices have it, and that is the part AI does well.
A short audit that costs an afternoon
Picture a restaurant owner with a walk-in cold room that a contractor fitted in 2018. She does not need to understand the software. She needs three answers from the contractor, in writing.
Is the controller reachable from the public internet? Does the maker still send it security updates? Can remote access be turned off, or limited to the contractor's own address?
If the answers are yes, no and no, the controller should be unplugged from the network until it is replaced, which is what Pearce did by instinct. Replacing it at the next service visit costs far less than a night of ruined stock. The same three questions fit a warehouse's loading-dock doors, a clinic's alarm panel or a school's cameras.
The attackers in these cases were not after any particular business, so the question is not whether anyone would want to hit you. It is whether the device can be reached, and whether anyone still updates it.