mundi-orch-intent-signal

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

• Third-party content exposure detected (high risk: 0.85). Perplexity web grounding (step 3) fetches public web content at runtime via mcp__perplexity__perplexity_research, and that retrieved free-form text is then used for classification/scoring, so outsider-authored prose can enter the agent’s LLM context indirectly.