Vulnerability  ·  2026-09-24

FastGPT (open-source LLM knowledge-base platform): SSRF guard bypass via DNS-rebinding TOCTOU in the safe-Axios interceptor (CVE-2026-84301)

VulnerabilityMedium impactGlobalCVE-2026-84301
NVD published CVE-2026-84301 (CVSS 6.3) on 2026-09-22: FastGPT's safe Axios request interceptor validates a hostname with isInternalAddress() before the HTTP connection performs an independent DNS lookup, enabling a DNS-rebinding TOCTOU SSRF bypass. Fixed in 4.15.2.
FastGPT RAG workflows fetch external content into the knowledge base; the SSRF bypass lets a crafted URL make the server reach internal infrastructure of the LLM platform deployment, which in cloud deployments can lead to metadata (credential) disclosure or internal service scanning.
The safe interceptor validates hostname via isInternalAddress() and then a later independent DNS lookup is used for the connection; an attacker flips DNS between validation and connection (DNS rebinding / TOCTOU) to have the server fetch internal addresses or cloud metadata despite the guard.
FastGPT (labring) < 4.15.2
Upgrade to FastGPT 4.15.2 (commit 0a38565c9d). Prefer resolving and validating the actual connected IP rather than a separate lookup; restrict outbound egress from FastGPT to deny internal/metadata ranges.
NVD CVE-2026-84301Fix commitcve.org record
See this in the live feed Explore related AI security and governance findings — updated every morning.
Open the feed →