FastGPT’s GET /api/core/ai/record/getRecord endpoint authenticates callers but retrieves LLM request and response traces solely using requestId, ignoring team scoping. This flaw enables any authenticated user who knows a requestId to read another team’s prompts, retrieved knowledge‑base chunks, and AI completions, thereby exposing proprietary or sensitive content generated by the AI platform. The vulnerability is identified as CWE‑639, a missing authorization in information disclosure.

labRing FastGPT versions prior to 4.15.0 that expose the /api/core/ai/record/getRecord API are affected. The issue is resolved in FastGPT 4.15.0. No other vendors or products are listed by the CNA.

The CVSS score of 7.1 indicates medium‑to‑high severity. The EPSS score of < 1% suggests a low but non‑zero likelihood of exploitation if an attacker obtains a valid requestId. The vulnerability is not included in the CISA KEV catalog, so no public exploits are known. The likely attack vector is an authenticated user sending a GET request to /api/core/ai/record/getRecord with a known or guessed requestId; no remote code execution or privileged escalation is involved, but confidentiality of LLM-generated data is compromised.