research

MEDIUM W011: Third-party content exposure detected (indirect prompt injection risk).

• Third-party content exposure detected (high risk: 0.85). The required workflow explicitly instructs searching public sources (e.g., Reddit/X/HN/StackOverflow/GitHub issues/forums/blogs/reviews/YouTube comments) and internal sources like Slack/CRM/support notes, which are outsider-authored free text that would be ingested into the agent’s LLM context via the runtime “search public sources” / “search internal sources” steps.