CVE-2026-77932 Details

A remote code execution vulnerability has been identified in DB-GPT versions through 0.8.2. The issue arises in the /api/v1/chat/react-agent endpoint, which relies on a mock authentication function that automatically assigns admin roles to all users, regardless of authentication. This flaw allows unauthenticated attackers to send prompts that trigger the ReAct agent's unsandboxed code_interpreter and shell_interpreter tools, executing arbitrary Python or bash commands as root within the webserver container. Additionally, this exploitation can access the bundled database using default credentials.

To address this vulnerability, replace the mock authentication function with a real authentication and authorization mechanism. Additionally, route the code_interpreter and shell_interpreter tools through the container runtime with appropriate network and capability restrictions, remove the LocalRuntime import from the shell_interpreter, and require authentication for the standalone dbgpt-sandbox API.