RAG Security is the newest and most specialized category in our taxonomy, focused on securing Retrieval-Augmented Generation pipelines — the architecture pattern where LLMs are connected to external knowledge bases to provide grounded, up-to-date responses. With only 3 tracked companies, this is an emerging category that we expect to grow significantly as RAG becomes the default architecture for enterprise AI applications.
The security challenges unique to RAG are substantial and underappreciated. Knowledge base poisoning — where attackers inject malicious content into the document stores that RAG systems retrieve from — can cause AI applications to generate harmful, misleading, or policy-violating responses while appearing to cite legitimate sources. Data exfiltration through retrieval — where carefully crafted queries cause the RAG system to surface and expose sensitive documents — represents a novel data loss vector that traditional DLP tools cannot detect.
Lasso Security, Noma Security, and Skyflow represent different approaches to this problem. Lasso focuses on monitoring and controlling GenAI interactions including RAG pipelines. Noma Security provides comprehensive AI application security including RAG-specific protections. Skyflow's data privacy vault can serve as a secure retrieval layer that enforces fine-grained access controls on the data flowing into RAG pipelines.
As RAG adoption accelerates across the enterprise — Gartner estimates that 60% of enterprise GenAI implementations will use RAG by 2027 — the security implications will demand dedicated solutions. We expect this category to expand significantly through acquisitions, new entrants, and feature extensions from adjacent categories (particularly LLM Application Security and AI Data Security). Early movers establishing expertise in RAG-specific threats will be well-positioned as the market matures.