External risk intelligence

DB-GPT Arbitrary File Write Via Skill Upload Leading to Code Execution

CVE advisorySeverity: CRITICAL (CVSS 9.3)

CVE-2026-80104

The vulnerability exists in an API endpoint used for uploading files that is accessible without authentication. As a web-based API service designed for agentic data management, such endpoints are intended to be network-reachable and often face the internet to facilitate external integrations or user interactions.

Path Traversal

Halo Surface Signal: 5 out of 5 — more likely to be public-facing.

External exposure likelihood

Horizon Alert

Summary of the vulnerability and why it matters

This CVE describes a vulnerability in DB-GPT that allows an unauthenticated remote attacker to execute arbitrary code on the server. The issue arises from how the system handles uploaded file paths, specifically allowing crafted filenames to write files to unintended locations on the server, potentially leading to the execution of malicious Python code.

  • Unauthenticated attackers can run code on the server.
  • Critical for systems handling sensitive data.
  • Confirm relevance and assess exposure.

Attack Path

How an attacker could exploit the issue

An attacker can exploit this vulnerability by sending a specially crafted filename during a file upload to write arbitrary content to any location on the server that the application's process has write access to. This could involve placing a malicious Python file within the application's directories. When this malicious file is subsequently imported by the application, it can lead to code execution on the server.

  • Network access required, no authentication needed.
  • Uploading a file with a specially crafted filename.
  • Arbitrary file write leading to code execution.

Live Threat

Current exploitation, exposure, and threat context

A remote attacker can upload files to arbitrary locations on the server, potentially overwriting existing files or introducing new malicious Python modules. This could lead to the execution of arbitrary code within the server process when these modules are imported by the application.

  • Server process code execution.
  • Arbitrary file write via crafted filename.
  • Server compromise and data exfiltration.

Operational Fix

Recommended remediation, mitigation, and detection steps

The Platform or Application Engineering team is likely responsible for managing DB-GPT deployments, while the Security Operations team should focus on identifying and isolating affected systems. The first practical step is to locate all instances of DB-GPT, determine their reachability and business criticality, and confirm the accountable owner for each deployment. Remediation efforts should then be prioritized based on this risk assessment, potentially involving vendor coordination for updates or implementing temporary risk reduction measures.

  • Platform/App Engineering owns the issue.
  • Verify network exposure and business criticality.
  • Plan vendor-coordinated remediation.

Supplementary metadata

Validate whether this threat affects your internet-facing exposure.

Halo Threat Intelligence helps prioritize remediation with Halo Surface Signal and H/A/L/O context. Start exposure validation with a free external attack surface trial.

Frequently asked questions

What is DB-GPT and how is it used?

DB-GPT is an open-source framework designed to help developers build applications powered by large language models and database integrations. It functions as an agentic data management platform, providing tools for data-driven tasks and workflow automation. Users deploy it to create intelligent systems that can process information and interact with structured or unstructured data sources via agent-based APIs.

How does CVE-2026-80104 allow code execution?

This vulnerability is a path traversal weakness, specifically categorized as CWE-22. Because the software fails to properly sanitize uploaded file names, an attacker can use special characters like parent references to bypass directory limits. By writing a malicious Python file into the application's own folders, the attacker tricks the system into treating their file as a legitimate part of the software, which then executes automatically when the application imports the fake module.

Do I need special privileges to trigger this bug?

No. The vulnerability exists within an API endpoint that performs an authorization check, but due to a flaw in how that check is implemented, it effectively grants admin-level access to anyone who sends a request. You do not need an account, a valid user ID, or any existing login session to successfully upload a crafted file. Requests sent without any credentials are treated with the same permissions as an authenticated administrator.

Is my DB-GPT instance at risk if it faces the internet?

According to Halo Surface Signal, this is highly likely to be a concern. Because this software is designed for agentic data management and external integrations, it is common for such API endpoints to be exposed to network traffic. If your instance is reachable from the internet, it is directly exposed to unauthenticated remote attackers who can exploit this path traversal flaw without needing to bypass perimeter defenses.

When should I take action to secure my deployment?

You should prioritize this immediately, as the flaw allows full control over the server process. Start by identifying all running instances of DB-GPT within your network and determining which are network-accessible. Coordinate with your platform engineering teams to verify if you are running an affected version. Once identified, restrict access to the affected API endpoints or coordinate with the software vendor for available updates to resolve the file path handling logic.

References