Source note: hidden document content and indirect prompt injection

Accessed 2026-08-17.

• OWASP GenAI Security Project: LLM01 Prompt Injection distinguishes direct and indirect prompt injection, notes that external files or webpages can influence model behavior, and recommends separating untrusted content, limiting privileges, and retaining human approval around higher-risk actions.
• Microsoft Support: Remove hidden data and personal information by inspecting files lists document properties, revisions, annotations, hidden text, and other non-obvious content. It specifically notes that white text on a white background is not detected as ordinary hidden text by Document Inspector.

These sources support a cautious workflow. They do not support a claim that one inspection method catches everything.

---

Source: labs/what-else-came-along/materials/sources.md