DB-GPT 0.8.2 Flaw Lets Attackers Run Code as Root Without Authentication
A critical vulnerability in DB-GPT versions up to 0.8.2 allows unauthenticated attackers to execute arbitrary code as root via a misconfigured API endpoint.
Key points
- CVE-2026-77932 is rated 9.3 Critical by the National Vulnerability Database.
- The flaw exists in the /api/v1/chat/react-agent endpoint of eosphoros-ai DB-GPT.
- Attackers can run Python or bash commands as root without authentication.
The National Vulnerability Database has assigned CVE-2026-77932 to a critical remote code execution flaw in eosphoros-ai DB-GPT. This vulnerability affects all versions of the application through 0.8.2. The issue allows unauthenticated attackers to execute arbitrary code with root privileges inside the webserver container.
At a glance
- CVE
- CVE-2026-77932
- Product
- eosphoros-ai DB-GPT
- CVSS
- 9.3 Critical
- Weakness
- missing authentication
- Exploited
- No known exploitation
The flaw stems from the /api/v1/chat/react-agent endpoint, which uses a mock get_user_from_headers function. This function incorrectly grants every caller the admin role. Consequently, attackers can send prompts that trigger the ReAct agent's unsandboxed tools. These tools include code_interpreter and shell_interpreter, which run arbitrary Python or bash commands.
Technical Details of the Vulnerability
According to the NVD record, the weakness is classified under CWE-306. The CVSS score is 9.3, which falls into the critical severity range. The vulnerability is unauthenticated, meaning no login credentials are required to exploit it. Attackers can leverage the ReAct agent's capabilities to bypass security boundaries.
The code_interpreter and shell_interpreter tools operate without sandboxing. This lack of isolation allows the execution of system-level commands. Attackers can run Python scripts or bash scripts directly within the container. The process runs as the root user, granting full control over the container environment.
Impact and Affected Systems
The affected software is eosphoros-ai DB-GPT, specifically versions up to and including 0.8.2. The vulnerability enables attackers to reach the bundled database. The database uses default credentials, which further simplifies access. Attackers can interact with the database directly from the webserver container.
This exposure allows for complete compromise of the container. The
What to do and how to stay safe
- Vulnerability severity scores ignore your specific architecture and context.
- Exploitability depends on network exposure and user behavior, not just code flaws.
- Prioritization requires continuous asset inventory and threat intelligence alignment.
Risk-based vulnerability management focuses your limited resources on the flaws that actually threaten your business. Start by mapping your critical assets and checking for active exploits to prioritize your backlog.
Step-by-step guide: Risk-Based Vulnerability Management: Prioritize Fixes by Actual Danger
General security guidance from the Threat Times newsroom. It is not confirmed advice from the organisations named in this story.
