CVE-2026-77932 is a critical unauthenticated remote code execution flaw in eosphoros-ai's DB-GPT. A mock authentication function gives every caller admin rights, so any attacker can make the ReAct agent run arbitrary Python or bash as root.
What Is It
DB-GPT through version 0.8.2 exposes the /api/v1/chat/react-agent endpoint. That endpoint uses a mock get_user_from_headers function, which assigns the admin role to every caller. Because of this, an unauthenticated attacker can send prompts that make the ReAct agent call its unsandboxed code_interpreter and shell_interpreter tools. Those tools run arbitrary Python or bash as root inside the webserver container. The attacker can also reach the bundled database, which uses default credentials.
The weakness is classified as CWE-306 (Missing Authentication for Critical Function). VulnCheck disclosed the issue, and it has been published in NVD. Its NVD status is currently "Deferred."
Why It Matters
- CVSS 3.1: 9.8 CRITICAL (
AV:N/AC:L/PR:N/UI:N/S:U/C:H/I:H/A:H) - CVSS 4.0: 9.3 CRITICAL (secondary score from VulnCheck)
An attacker needs only network access. No privileges or user interaction are required, and attack complexity is low. A successful attack gives root-level code execution in the webserver container, with high impact to confidentiality, integrity and availability. Default database credentials widen the impact past the application layer.
The supplied CISA KEV data contains no entry for this CVE, so active exploitation is not confirmed by KEV at this time. The CVSS 4.0 exploit maturity field is "Not Defined."
What's Vulnerable
- Vendor: eosphoros-ai
- Product: DB-GPT
- Package:
pkg:pypi/dbgpt-app - Affected versions: all versions from 0 through 0.8.2 (inclusive)
- Repository: https://github.com/eosphoros-ai/DB-GPT
The NVD references point to the mock authentication logic in dbgpt_serve/utils/auth.py and the code interpreter tool in dbgpt_app/openapi/api_v1/tools/code_interpreter.py.
Patch Status
The supplied NVD and KEV data does not name a fixed version, and it lists no CISA required action or due date. The data marks every version through 0.8.2 as affected. Operators should check the vendor repository and the VulnCheck advisory for remediation guidance. They should also treat any DB-GPT instance reachable from the network as exposed until a fix is confirmed.