Embed in a route
<FigureFromSource sourcePath="content/publications/figures/products/rag-harness.dsl.md" figureId="RAG-02" />
Add selected Workbench capabilities through bounded OEM and partner integrations
RAG-02
A useful harness separates prompt injection, corpus poisoning, cross-tenant retrieval, provenance loss, and unsafe action propagation.
coverage matrix
Indirect prompt injection
Hostile instructions enter through retrieved content.
Corpus poisoning
Malicious or misleading content affects retrieval and generation.
Cross-tenant retrieval
Content crosses an intended tenant or access boundary.
Provenance loss
The system cannot preserve or communicate source origin and trust.
Unsafe action propagation
Retrieved content influences a tool or consequential action.
Hostile instructions enter through retrieved content.
Malicious or misleading content affects retrieval and generation.
Content crosses an intended tenant or access boundary.
The system cannot preserve or communicate source origin and trust.
Retrieved content influences a tool or consequential action.
This coverage matrix frames a RAG harness around five failure modes that commonly break end-to-end reliability: indirect prompt injection, corpus poisoning, cross-tenant retrieval, provenance loss, and unsafe action propagation. It maps those risks across the full pipeline, from ingest and retrieve through assemble, generate, act, and retest, so each stage can be evaluated for both attack exposure and control effectiveness.
The abstract emphasizes that useful coverage is not just about finding failures, but about making them observable and repeatable. By assigning states such as covered, partial, planned, unknown, and not applicable, the harness reveals where defenses are already exercised, where evidence is weak, and where the system still permits risky behavior to propagate into downstream actions.
<FigureFromSource sourcePath="content/publications/figures/products/rag-harness.dsl.md" figureId="RAG-02" />
RAG Failure-Mode Coverage (RAG-02). AI Security LLC Figure Library. https://aisecurity.llc/publication-dsl/figures/RAG-02