Skip to content

Security starts in hardware

The physical layer is the part of the security stack nobody inspects. Every audit reaches the algorithms, the configuration and the process, and stops above the silicon that generates the keys. What changes when that layer becomes measurable is not that it gets stronger — it is that it becomes arguable.

Security reviews have a floor. They examine architecture, code, configuration, key management and process, and somewhere near the bottom they reach a component with a datasheet and stop. Below that line the questions change from "show me" to "the vendor says".

That floor sits in an odd place. It is above the layer that decides whether the keys were unpredictable.

The inversion

The upper layers of the stack are the most inspected and the most replaceable. A misconfigured cipher suite is found by a scanner and fixed in an afternoon. A vulnerable library is flagged by a dependency check and patched.

The lower layers are the least inspected and the least replaceable. A device identity key generated from a weak source at first boot cannot be fixed by any update, because the update is authenticated with it. The scrutiny is inversely proportional to the cost of being wrong.

Why it stayed that way

Not negligence. For most of the history of this discipline there was nothing to inspect. An entropy source was a component with a specification, and the only available evidence was that it had been evaluated once, by someone else, on a different unit.

You cannot audit what does not report. So the profession developed the reasonable habit of treating the physical layer as an assumption, and built increasingly sophisticated analysis on top of an input nobody measured.

What changes when the layer becomes measurable

Not that it becomes stronger. A source that reports its health is not more unpredictable than one that does not.

What changes is that the assumption becomes a claim, and a claim can be checked. The question "how do you know your entropy is good" moves from unanswerable to answerable, and once a question is answerable, someone eventually asks it.

That has already happened one layer up. Software supply chains were an assumption until build provenance made them checkable, and then within a few years they became a requirement. The mechanism was not new fear — it was that an answer became possible.

The uncomfortable part

If the physical layer becomes measurable, most deployed systems will measure badly. Not because their hardware is poor, but because they have no way to say anything about it at all — the source most systems actually use is documented well enough to integrate and not well enough to evaluate.

That implicates the installed base rather than any one competitor, which makes it an awkward thing to say and an accurate one.

What we are actually arguing

Not that software security is weak, or that hardware is a substitute for it. Most breaches are still logic errors, stolen credentials and misconfiguration, and no entropy source addresses any of those.

The argument is narrower and, we think, harder to dismiss: there is one layer of the stack that everything else depends on, that nobody currently inspects, and that has recently become inspectable. Making it inspectable is worth doing before someone asks and the answer is a shrug.