Skip to content

Entropy vs randomness, a distinction that matters

Randomness describes how output looks; entropy measures how much unpredictability is actually present. A stream can be perfectly random-looking and contain no entropy at all — which is precisely what a good pseudorandom generator produces, and why statistical testing cannot establish security.

The distinction in one example

Encrypt a simple counter under a fixed AES key. The output is uniform, uncorrelated and passes every statistical suite ever written. Its entropy, to anyone holding the key, is zero. Nothing about the bits reveals this.

That is the whole distinction. Randomness is a description of appearance. Entropy is a quantity describing what an observer does not know. Cryptography needs the second, and can only measure the first directly.

Entropy is relative to an observer

This is the part that trips people up. Entropy is not a property the bits carry around with them. It is a measure of uncertainty relative to what someone knows. The counter stream has zero entropy for the key holder and looks fully random to everyone else.

So "how much entropy does this data have" is not answerable without saying who is asking and what they already know. In practice the observer is the attacker, and the model of what they know is the assumption the estimate rests on.

Why the loose usage causes real harm

When the words are used interchangeably, a test result gets mistaken for evidence of security. A team runs a statistical suite, the source passes, and everyone concludes the randomness is fine. What was established is that the output has no gross statistical defect — useful for catching a broken source, and silent on unpredictability.

Statistical tests are not incoming inspection takes that further — it is the most common misunderstanding in this field.

The vocabulary