Executive brief
FastGPT is an AI agent building platform. A security flaw in its tool-testing feature allows anyone on the internet to use the server as a proxy to access internal systems. This could lead to the theft of sensitive AI provider API keys (like OpenAI or DeepSeek), exposure of internal database information, and unauthorized access to other private services running on the same network.
Technical details
The FastGPT HTTP tools testing endpoint (/api/core/app/httpTools/runTool) was exposed without any authentication middleware. This endpoint functions as a full HTTP proxy, accepting user-supplied parameters including baseUrl, toolPath, HTTP method, custom headers, and body. Because FastGPT is commonly deployed via Docker Compose, an unauthenticated remote attacker can use this endpoint to perform Server-Side Request Forgery (SSRF) against internal services. This allows for the exfiltration of third-party API keys from the AI Proxy management API, retrieval of MongoDB diagnostic data, and access to internal verification codes. The vulnerability has been addressed in version 4.14.9.5 by enforcing authentication and implementing internal address checks.
Affected products
- labring FastGPT < 4.14.9.5
Timeline
- 2026-03-25: patched: Fix merged in pull request #6640 and released in v4.14.9.5
- 2026-03-26: advisory: GitHub Security Advisory GHSA-w36r-f268-pwrj published
- 2026-03-31: disclosed: CVE-2026-34162 published to NVD