Db Gpt EUVD-2026-13806

This vulnerability presents moderate-to-high real-world risk despite the CVSS 6.3 score (medium severity). The CVSS vector (AV:N/AC:L/PR:L/UI:N/S:U) indicates network-accessible attack surface requiring low complexity and authenticated access, with limited scope and moderate impact (confidentiality, integrity, availability). However, several factors elevate practical risk: (1) a public proof-of-concept exploit exists and is disclosed on GitHub, eliminating the need for attacker sophistication; (2) file upload vulnerabilities frequently chain into remote code execution, escalating impact beyond the base CVSS; (3) the vendor's non-responsiveness to early disclosure suggests no official patch timeline; (4) authentication is required, limiting attack surface but not eliminating risk in environments with multiple users or weak credential management. The exploitation is highly automatable given the POC availability. This should be prioritized for immediate patching or mitigation, particularly in production environments handling sensitive data.

An authenticated user (such as a legitimate employee with credentials) or an attacker who has compromised valid credentials connects to the DB-GPT FastAPI endpoint and sends a crafted HTTP POST request to the module_plugin.refresh_plugins endpoint with a malicious file (e.g., a Python script or shell command) embedded in the upload payload. Due to insufficient validation, the file is written to the server's filesystem in a location where it can be executed (such as a plugin directory with automatic loading). Upon execution, the attacker gains arbitrary code execution on the DB-GPT server, potentially compromising the underlying database, exfiltrating data, or pivoting to adjacent systems. A public POC demonstrates this exact technique, making exploitation straightforward for attackers with valid credentials.