Sensitive Imagery (LLM-as-Judge)
This is not an attack detector — it's a content-review check that asks an AI model to judge whether an image passed through an AI agent contains sensitive or potentially harmful material. It flags for human review only; it never blocks anything.
How the attack works
An image enters the system, for example as input to or output from an AI agent. The rule sends that image to an AI scanner along with a rubric asking whether the content is sensitive or harmful. The scanner returns a judgment, and the rule logs it as a report-level finding. No action is taken beyond logging — the pipeline continues.
Netzilo detection
Netzilo reports this behaviour when it is observed.
- Signature ID
- netzilo-llmjudge-sensitive-imagery-001
- Severity
- Medium
Why it matters
On its own this rule does not prevent harm; it only produces a record that a human or downstream process can review. If nobody looks at the reports, sensitive imagery can pass through unnoticed. Organizations relying on this as their only safeguard have no actual blocking control in place.
What you can do
- →Treat this as a monitoring signal, not a control — pair it with an actual blocking policy if you need to stop sensitive images, not just log them.
- →Route flagged reports to someone who reviews them regularly rather than letting them pile up unread.
- →Verify your AI scanner is actually configured and reachable, since this check fails open (allows everything) if no AI keys are set.
- →Expect false positives on legitimate security research, documentation, or test imagery that discusses sensitive topics without being harmful, and build a quick review/override process for those cases.
Known benign look-alikes
- Legitimate content discussing the flagged topic (security research, documentation, test data)
- AI scanner unavailable or no AI keys configured (fails open -> allow)