Covert Channel

A covert channel — the term is also used in German as “verdeckter Kanal” — is a communication path within a computer system that was never actually meant to carry information. Normal communication runs through official routes: a file is saved, a message is sent, a network packet is transmitted. A covert channel instead takes a detour — for example, how long an operation takes or how much memory is currently in use. The system notices nothing, because it doesn’t recognize these signals as communication at all. The term comes from computer security and describes a class of vulnerabilities that are especially hard to detect.

Why covert channels undermine security architectures

Modern operating systems carefully separate programs and users from one another. One process is not allowed to simply read another’s data. This separation is called isolation. Covert channels subvert exactly this isolation — not by breaking the protective mechanisms, but by going around them.

That is what makes them truly dangerous: an attacker doesn’t need a security flaw in the classic sense. All that’s needed is two processes sharing a common resource — such as the processor core or the chip’s cache memory. Through the behavior of that resource, bits can be transmitted without any security software ever raising an alarm. The channel is invisible because it was never designed as a channel in the first place.

Two basic types and how they work

Broadly speaking, two types are distinguished. In a storage-based covert channel, two programs write to and read from a jointly accessible resource — for example, taking turns using a file whose mere existence or size carries the signal. In a time-based covert channel — also called a timing channel — the signal is time itself. Process A performs an action or withholds it; Process B measures how long it has to wait for a shared resource. A short wait means “0,” a long wait means “1.” This produces a data stream made up of nothing but timing differences.

A well-known real-world example is the Spectre vulnerability from 2018. It exploited a timing channel in the processor core: an attack program could read portions of other programs' memory by measuring how quickly the chip accessed certain data. Spectre affected nearly all modern processors worldwide — Intel, AMD, and ARM. The vulnerability was so hard to close because it wasn’t a flaw in software code, but was built into the fundamental design of the chip itself.

Covert channels also show up in AI systems. Researchers have shown that a language model running inside a so-called sandbox — an isolated environment — can embed hidden information within its visible responses. For instance, through the choice of specific words or the length of sentences. This is an active area of research, because such channels are difficult to prove and even harder to prevent.

Where covert channels show up in practice

In IT security, the term comes up above all in connection with high-security systems. Government agencies or companies that process top-secret data must ensure that no covert channel exists through which data could leak to the outside. This is referred to as an “air-gapped” system: a computer with no network connection. But even then, covert channels have been demonstrated — for instance via the noise level of a cooling fan or fluctuations in the power supply.

The topic regularly comes up in tech news whenever new hardware security vulnerabilities are discovered. After Spectre came similar attacks with names like Meltdown, Rowhammer, or Hertzbleed — all exploiting variants of covert channels. Chip manufacturers and operating system developers then often have to intervene deep within the architecture to close off these channels. This frequently costs computing performance, because deliberate imprecision has to be built in so that an attacker can no longer make reliable timing measurements.

For the AI debate, the term becomes relevant when discussing autonomous systems meant to run in isolated environments. The question of whether a model sends hidden signals intentionally or by accident is hard to answer — and that is exactly what makes covert channels one of the most fascinating and uncomfortable topics in modern computer security.