Executive brief
BISHENG is an open-source LLM DevOps platform for enterprise AI applications. An unauthenticated attacker can exploit a flaw in the workflow callback API to make the server request arbitrary URLs, allowing them to probe internal network services (including cloud metadata endpoints) and steal captured responses from cloud storage. This could expose internal systems and sensitive cloud credentials.
Technical details
The vulnerability is a server-side request forgery (SSRF) in the POST /api/v1/workflow/report/callback endpoint. The endpoint lacks authentication checks and does not restrict URL schemes or filter hostnames, allowing unauthenticated attackers to supply arbitrary URLs. When invoked, the server fetches those URLs and stores responses in object storage. An attacker can enumerate internal network services and cloud metadata endpoints (e.g., AWS IMDSv1), then retrieve the captured responses by supplying caller-controlled object names. No patch availability is stated in the advisory.
Affected products
- dataelement BISHENG through 2.6.0-fix2
Timeline
- 2026-08-28: disclosed