How we think about attack surface

Sample post. The reasoning here is deliberately generic and names nothing real.

Most of research is choosing what not to look at. Attention is the scarce resource, so the first question is never “is this exploitable?” but “is this worth an afternoon?” Here is the shape of how we tend to decide.

Start from the boundary

A bug only matters if it crosses a boundary someone cares about. Before reading any code we try to name the boundary in one sentence: who is on each side, and what one side is trusted to assume about the other. If we cannot write that sentence, the target is not ready to audit yet.

Prefer reach you can measure

We rank candidates by two rough keys:

  1. Reach - who can get to the code, and what they must already hold to do it.
  2. Ceiling - the best outcome if the worst case is true.

A defect with modest reach that any user can trigger usually beats a dramatic one that needs a setup nobody has. When reach is still a guess, we say so rather than dressing it up.

Keep a written trail

Every session ends with a note, even a dead end. A null result is a claim about our method as much as about the target, so we record what we actually looked at. The cheap discipline of writing it down is what keeps us from re-walking the same ground a month later.

None of this is a formula. It is just a way to spend a limited budget on the questions most likely to pay.

All posts