About Services Pricing Case Studies Blog Contact Book Free Audit →
📊 LIVE INDUSTRY BENCHMARK

State of Agent Security

A real-time, data-driven look at the vulnerability landscape for AI agents, LLMs, and MCP. Tracked continuously from CISA KEV and NVD feeds, filtered for what actually matters to teams shipping agents in production.

Data updated daily · 6:00 AM CT
709
AI/Agent CVEs (14-day)
1,660
CISA KEV total
51
KEV AI-relevant
3,742
NVD CVEs (window)

Where the Risk Is Concentrating

AI/agent-relevant CVEs broken down by category over the tracking window.

RAG / Retrieval
171
171
MCP / Tool / Agent
73
73
LLM / Foundation Model
26
26
Prompt Injection
6
6
Generative AI / Chatbot
10
10

What the Data Says

RAG is the new front line.

Retrieval-augmented generation vulnerabilities now dwarf every other agent-security category in our feed. If you connect an agent to a vector database or knowledge base, that retrieval layer is the most likely place an injection lands.

MCP is an emerging, under-reported surface.

Model Context Protocol — the standard connecting agents to tools — is showing up as a distinct CVE category. Every tool your agent can call is a potential privilege boundary, and MCP tool-abuse vectors are the ones teams most often miss.

Known-exploited counts are the canary.

51 AI/agent-relevant entries sit in CISA's known-exploited catalog. When a vulnerability is in KEV, it's being actively exploited in the wild — not theoretical. That's the number to watch for urgency.

Latest Relevant CVEs

CVEDateCategory
CVE-2026-706192026-08-04RAG
CVE-2026-455382026-08-04MCP / Agent
CVE-2026-704772026-08-04Prompt Injection
CVE-2026-704782026-08-04LLM
CVE-2026-704922026-08-04RAG
CVE-2026-679792026-08-04RAG
CVE-2026-476822026-08-04RAG
CVE-2026-704742026-08-04RAG
CVE-2026-704862026-08-04RAG
CVE-2026-704852026-08-04RAG

Full dataset tracked locally. This page shows the most recent high-signal entries.

Methodology

Want to know where YOUR agent stack stands?

This is the industry picture. The only way to know your own exposure is to test it — against the same attack classes driving these CVEs.

Book a Free 15-Min Audit → Or run your system prompt through the free AI System Prompt Scanner