sap-ai-core
Pass
Audited by Gen Agent Trust Hub on Sep 15, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADS
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill facilitates the creation of systems that ingest untrusted data from document repositories (Grounding) and process user-supplied chat history, which are standard surfaces for indirect prompt injection. However, it also provides detailed instructions for implementing security controls.\n
- Ingestion points: Grounding pipelines fetch data from Microsoft SharePoint, AWS S3, and SFTP servers (
references/grounding-rag.md). Chat history management also ingests previous conversation turns (references/advanced-features.md).\n - Boundary markers: The orchestration module mandates a templating structure with defined roles (system, user, assistant), establishing clear boundaries for model instructions.\n
- Capability inventory: The platform supports model inference and batch execution via Argo Workflows.\n
- Sanitization: The skill provides extensive documentation on mitigating risks using Azure Content Safety, Llama Guard 3 for content filtering, and Data Masking for PII protection (anonymization and pseudonymization).\n- [EXTERNAL_DOWNLOADS]: The skill references official vendor-maintained Python packages for extending core functionality.\n
- Evidence: Mentions
sap-ai-core-datarobotandsap-computer-vision-packagefor integration tasks.\n - Context: These are legitimate vendor resources used for development within the SAP ecosystem.
Audit Metadata