The vulnerability arises from a missing authorization check in the promptExperiment handler of AiragPromptsController, allowing any authenticated user to execute AI prompt experiments. This can lead to low‑privileged attackers supplying other users’ prompt template and dataset identifiers, triggering large language model evaluation runs, writing results into airag_ext_data, and modifying dataset status. The flaw is a direct authorization bypass (CWE‑862), resulting in unauthorized access to application functionality and data integrity.

JeecgBoot version 3.9.5 and earlier are affected. This impact applies to all deployments of the JeecgBoot application that include the AiragPrompts module, specifically the /airag/prompts/experiment endpoint.

With a CVSS score of 5.3, the vulnerability presents a moderate risk; no EPSS data is available and the flaw is not listed in the CISA KEV catalog. Any user who can authenticate to the application can exploit the missing authorization guard. An attacker with minimal privileges could launch arbitrary AI prompt experiments, correlate with other users’ data, and produce or alter results, thereby compromising data confidentiality, integrity, and potentially affecting downstream processes.