medbeads-agent-native-immutable-data

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

• Third-party content exposure detected (medium risk: 0.65). The runtime workflow “Serialize the DAG subgraph for LLM prompt injection” ingests content from Beads created from FHIR bundles (and potentially external evidence URIs), so if those Beads originate from outsider-authored patient records/notes, their free-text fields (e.g., medical_note, content payloads) are serialized into the LLM prompt as context.