data-monitoring

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

• Third-party content exposure detected (high risk: 0.75). The required workflow includes “Mandatory Search First” and “Every Claim Must Have a Source,” implying the agent will fetch and read public web pages at runtime (outsider-authored free text) to produce benchmarks and attach search records into the LLM context/output.