MCP Fetch Server SSRF (CVE-2026-104120) Has a Public Exploit and No Fix
CVE-2026-104120 lets an agent steer the official MCP fetch tool to internal and cloud metadata addresses. A public exploit exists; fix PR #4890 is unmerged.
· 6 minDesk · AI systems and model risk
Coverage of AI application security, model abuse, agentic systems, and data exposure.
CVE-2026-104120 lets an agent steer the official MCP fetch tool to internal and cloud metadata addresses. A public exploit exists; fix PR #4890 is unmerged.
· 6 minn8n's queue-mode handler skipped name, permission, checksum and npm safety checks, so Redis write access could install any npm package cluster-wide. Fixed; no known exploitation.
· 5 minA langgraph-sdk bug applied actions=-scoped auth handlers to every action, so an authenticated user could reach others' threads. Fixed in 0.4.4; what to check.
· 6 minThree GitHub advisories trace how Vibe-Trading's open API led to a root shell and stolen broker, LLM and cloud keys. Fixed in 0.1.7; upgrade to 0.1.16.
· 8 minGoogle's GTIG logged 141 exploited flaws in Jan–Aug 2026, more than all of 2025, and lists LiteLLM CVE-2026-42271 and Langflow CVE-2026-5027 as exploited.
· 8 minLightLLM through 1.2.0 runs an unauthenticated, pickle-enabled RPyC service on visual_only nodes. Anyone who can reach the port can run code, and no fix exists yet.
· 6 minGitLab fixed a CVSS 9.9 prompt-template sandbox escape in its AI Gateway. Self-hosted gateways must move to 19.2.4, 19.3.2 or 19.4.1.
· 7 minObot 0.21.1–0.24.1 skips its ACR check on the composite MCP route, so any signed-in user can reach restricted servers. Obot says v0.25.0 fixes it.
· 5 min