research-icp

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

• Third-party content exposure detected (high risk: 0.85). Outsider free text is ingested via the runtime VoC collection path: voc-collector-agent fetches and extracts public web/community content (e.g., Reddit/G2/Twitter reviews/threads) and turns it into readable quote text that is then placed into the LLM context for downstream synthesis/critic steps.