FastGPT, an open‑source AI knowledge base platform built by labring, contains a flaw in its POST /api/core/dataset/collection/create/reTrainingCollection endpoint from versions 4.14.17 up to, but not including, 4.15.0-beta4. An authenticated tenant user can invoke this endpoint with a server‑owned datasetId belonging to another tenant, causing the service to persist a mixed‑tenant dataset object. The downstream dataset, collection, and training APIs then perform authorization checks against these inconsistent ownership anchors, potentially granting read, update, or delete privileges over data owned by other tenants. This vulnerability is classified as CWE‑915: server‑side data‑fencing bypass.

The affected product is FastGPT from labring. Versions 4.14.17 through the end of the 4.14 series contain the vulnerability; the issue was fixed in version 4.15.0-beta4 and later.

The CVSS score of 6.3 indicates moderate severity, and the EPSS score of less than 1% indicates a low likelihood of exploitation in the wild. The vulnerability is not listed in the CISA KEV catalog. An attacker must be an authenticated tenant user and must know the identifiers of target objects to exploit the flaw; no other conditions are required.