Assess a whole application's attack surface, on your own lab and on paper

capstone · 150 min · Objective 4.6

Task

Bring the whole of Domain 4 together: take an application you built with an LLM-backed feature, exploit the conventional attack surface hands-on in your own lab, and assess the AI feature's exposure on paper — producing one attack-surface document that a client could act on. This is the attack domain's reasoning, end to end.

Steps

  1. On the lab you own, run an application you built with a conventional surface (a web form, an authentication flow, a file handler) and an LLM-backed feature.
  2. Hands-on on your own app: find and prove one conventional flaw from earlier in the domain — injection, traversal, an authentication weakness — read-only and without destroying data.
  3. On paper for the AI feature: assess where untrusted input reaches the model, direct and indirect injection, and what the feature can reach if steered.
  4. Combine both into /tmp/surface.md: the conventional findings with evidence, and the AI exposure assessment, each with impact for this application.
  5. Write the combined attack narrative: how a foothold on the conventional side and the AI feature's reach relate.
  6. Write remediation for each, at the level of the cause.

Verify

grep -cE "^##\s+(Conventional findings|AI exposure|Narrative|Remediation)" /tmp/surface.md
grep -cE "REDACTED|\*{3,}|<masked>" /tmp/surface.md
grep -ciE "no model|on paper|assessed|not attack" /tmp/surface.md
grep -ciE "impact.*this application|reaches|blast radius" /tmp/surface.md

The first count must be at least 4 — the document has its required sections. The second must be non-zero: the conventional evidence masks real values. The third must be non-zero: the AI feature was assessed on paper, not attacked. The fourth must be non-zero: every finding carries impact for this application. That combined document is Domain 4's argument in one deliverable.

Notes

Attacks and Exploits is 35% of PT0-003, and this capstone is why it is not just a list of techniques: the conventional surface is exploited hands-on on your own app, read-only, while the AI surface is assessed on paper because CompTIA scopes it as explaining rather than performing. The combined narrative is what turns individual findings into the real risk a client faces. Everything hands-on ran on an app you built; no model was attacked.

This is an independent study companion for CompTIA PenTest+ PT0-003 and is not produced by or endorsed by CompTIA.