Everyone is worried about AI security. But who is protecting the AI?

Most conversations about AI and security start in roughly the same place; cyber risk, deepfakes, fraud and alike…

The assumption is usually that AI is the thing creating the security problem. But AI has another problem of its own.

Someone has to protect it.

For all the language of clouds, models and algorithms, AI is still built on very physical things, and expensive things.

Semiconductors, cooling equipment, copper, fibre. Even the trucks carrying equipment are worth millions of dollars.

Criminals have noticed.

WIRED recently reported on increasingly violent cargo thefts targeting high-value AI hardware in the US, including incidents where thieves allegedly used deliberate vehicle collisions and PIT-style manoeuvres to disable security escorts and steal shipments of data-centre equipment.

The story is remarkable. The underlying logic is not.

Crime tends to follow value. AI has spent the last few years creating an extraordinary amount of it.

AI depends on highly specialised semiconductors, servers and data-centre infrastructure that can cost millions before they have even been installed. That makes them attractive targets.

Fortune recently reported that organised cargo theft is increasingly targeting servers, chips and other technology destined for AI infrastructure. Citing CargoNet figures, it reported that cargo theft losses reached around $725 million in 2025, while the average value of individual thefts rose significantly.

Cargo theft itself is hardly new. What changes is the value profile.

The more expensive, scarce and strategically important a component becomes, the more attractive it becomes to organised criminals.

AI hardware is increasingly ticking all three boxes.

The equipment has been manufactured, stored, moved between facilities. Loaded and transported, several times over.

The physical attack surface therefore extends across the entire supply chain and creates opportunities for more than traditional theft.

It creates opportunities for identity fraud, insider involvement, shipment diversion and organised groups deliberately targeting high-value routes.

The WIRED reporting is particularly striking because some thieves are no longer simply looking for an unattended trailer.

They are reportedly willing to attack the security operation protecting it.

At that point, AI infrastructure has stopped looking like an IT-security issue.

It looks much more like the protection of any other high-value critical asset.

The Wall Street Journal recently reported that Taiwanese prosecutors charged nine people over an alleged attempt to illegally ship 130 advanced AI servers to China using false information about their intended destination.

That changes the nature of the asset again.

A server can simultaneously be: commercially valuable; attractive to organised crime; subject to export restrictions; and strategically important.

The security question is therefore no longer simply: How much is this equipment worth?

It may also be: Who wants it, and why?

Technology does not stop being physical simply because the user cannot see the machinery. Somewhere, an AI model is running on a server.

That server contains expensive chips. Those chips travelled through a supply chain.

The building they operate from depends on power, cooling and network infrastructure.

And every stage creates something that can be stolen, disrupted, damaged or deliberately targeted.

Security teams have seen this pattern before. As an asset becomes more valuable, the effort required to protect it usually rises with it.

The only unusual thing about AI may be the speed at which that value has been created.

We are spending enormous amounts of time asking how AI might change security. Perhaps the reverse question deserves a little more attention.

As AI becomes more valuable, more strategic and more physically widespread, how will security operations have to change to protect it?