AI-Powered RAG Workflow for n8n — agentic threat model

Not certain from the listing — The specific foundation LLM is not disclosed. Standard LLM risks apply, including prompt injection that could bypass system instructions to exfiltrate retrieved document context or generate misaligned outputs.

The workflow relies on Google Drive and Pinecone. Key threats include data poisoning (uploading malicious files to Google Drive to corrupt the vector database), embedding inversion attacks, and the lack of document-level access control mapping from Google Drive to Pinecone, potentially exposing restricted files to unauthorized users.

Built on the n8n workflow automation framework. Threats include insecure tool integration, workflow injection vulnerabilities where malicious inputs manipulate n8n node execution, and unauthorized execution of downstream n8n nodes if the workflow is poorly scoped.

Not certain from the listing — The deployment model (n8n Cloud vs. self-hosted) is unspecified. Threats include exposure of n8n webhook endpoints, insecure storage of Google Drive OAuth tokens and Pinecone API keys within the n8n credentials manager, and lack of network isolation.

Not certain from the listing — There is no mention of built-in evaluation, guardrails, or logging mechanisms. This creates blind spots regarding prompt injection attempts, data leakage, and drift in retrieval accuracy.

Not certain from the listing — Authentication relies on standard Google OAuth and Pinecone API keys, but there is no evidence of fine-grained access controls, audit logging, or compliance mapping (e.g., GDPR/CCPA) for data stored in the vector database.

Not certain from the listing — While n8n allows multi-app integration, this is described as a single-agent RAG pipeline. The primary ecosystem threat is cascading failures or unauthorized data sharing if the workflow is connected to other active agents or APIs.