Public AI agent threat intelligence
See what threatens autonomous systems.
Search evidence-backed threats, vulnerabilities, malicious packages, and campaigns across MCP, OpenClaw, and Hermes.
- 27
- Verified records
- 0
- Active exploitation
- 0
- Unpatched
- Oct 16, 2026Not a full-database recheck
- Latest record verification (UTC)
Evidence first
Threat index
| Score | Threat | Ecosystem | Status | Verified |
|---|---|---|---|---|
| 55Medium risk | LangChain serialization can expose secrets through untrusted agent dataGHSA-c67j-w6g6-q2cm · CVE-2025-68664 · DATA EXTRACTION | Patched | ||
| 53Medium risk | MCP loopback could expose owner-only tools to non-owner runsGHSA-52xj-c9p8-78cv · OPENCLAW EXPLOIT | openclaw + mcp | Patched | |
| 50Medium risk | Feishu tools could ignore per-account disablementGHSA-2q7j-2vhx-56g8 · OPENCLAW EXPLOIT | openclaw | Patched | |
| 50Medium risk | Feishu permission tools could ignore per-account disablementGHSA-w8wf-3qvj-6xqf · OPENCLAW EXPLOIT | openclaw | Patched | |
| 50Medium risk | flock wrapper could bypass durable exec approval bindingGHSA-3fp5-v549-9v66 · OPENCLAW EXPLOIT | openclaw | Patched | |
| 50Medium risk | Discord guild actions could skip cross-provider requester authorizationGHSA-3pmr-x9g8-m55r · OPENCLAW EXPLOIT | openclaw | Patched | |
| 50Medium risk | Browser CDP discovery could accept blocked WebSocket URLsGHSA-3x84-qq85-fj65 · OPENCLAW EXPLOIT | openclaw | Patched | |
| 50Medium risk | Sandbox bind mounts could bypass parent-directory denylist checksGHSA-575v-8hfq-m3mc · OPENCLAW EXPLOIT | openclaw | Patched | |
| 50Medium risk | Plugin install commands could allow non-owner persistenceGHSA-7vrr-rp4x-4g76 · OPENCLAW EXPLOIT | openclaw | Patched | |
| 50Medium risk | Host exec environment filtering could allow Git ext transportGHSA-9969-8g9h-rxwm · OPENCLAW EXPLOIT | openclaw | Patched |