magic-linguistic-eval

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

• Third-party content exposure detected (high risk: 0.72). Step 2/4/5/6 of the workflow ingests outsider-authored benchmark and contamination-audit reference text (e.g., FLORES/NTREX/Belebele/AfroBench/SEACrowd and their dataset contents) into the LLM context via the eval report generation and metric/probe orchestration, which are not authored by the operating user.