tetlock

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

• Third-party content exposure detected (high risk: 1.00). The skill explicitly instructs spawned teammates to perform WebSearch/WebFetch on public sites (see the Calibrator: "Use WebSearch to find actual data"; the Updater: "Use WebSearch to find the most relevant current evidence"; and the Scorekeeper: "Use WebSearch and WebFetch" and to pull prediction market prices), and those retrieved third‑party pages are read and used to update probabilities and drive decisions, so untrusted web content can materially influence agent behavior.