Pre-Ingestion Scanning Pipeline

Documents are scanned for injection markers, malware, and policy violations before embeddings are written.

Impact

Stops poisoned or instruction-laden content from entering the retrieval corpus.

Why this severity

Guardrails take the highest severity of the risks they mitigate: Indirect Prompt Injection via Documents (Critical), Vector DB Poisoning (High), Lack of Context Sanitization (High).

Method: OWASP Risk Rating Methodology

Mitigates

Indirect Prompt Injection via Documents Threat · Critical severity
Vector DB Poisoning Threat · High severity
Lack of Context Sanitization Vulnerability · High severity

Needed by

Document-Level Access Enforcement Guardrail · Fixes high risks

Remediation

Run static detectors for instruction phrases and hidden text layers.
Quarantine failures into a review queue instead of indexing.
Re-scan on every document version bump.

How to verify

Known-bad fixtures never reach the production index.
Quarantine SLAs and reviewer audit trails are intact.

Pattern

const report = await scan(doc);
if (report.risk >= 'HIGH') return quarantine(doc, report);

Framework mapping

OWASP Top 10 for LLM Applications 2025

LLM01:2025 Prompt Injection
LLM04:2025 Data and Model Poisoning
LLM08:2025 Vector and Embedding Weaknesses

MITRE ATLAS

AML.M0030 Restrict AI Agent Tool Invocation on Untrusted Data
AML.M0007 Sanitize Training Data

NIST AI RMF 1.0

MEASURE 2.7 AI system security and resilience are evaluated and documented