models-roundtable

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

• Third-party content exposure detected (medium risk: 0.30). The skill’s Phase 1 “Blind Fan-out” reads the same raw prompt supplied by the user (the “raw prompt verbatim” inserted into the seat prompt), and that text is ingested directly by every seat/organizer/judges/synthesizer as part of the record at runtime.