Executive brief
UpTrain is an open-source platform used to evaluate and improve generative AI applications. The `/create_project` endpoint contains a vulnerability that allows authenticated users to execute arbitrary code on the host system running UpTrain. In typical deployments, this means an attacker could take complete control of the Docker container hosting the application.
Technical details
The vulnerability is a code injection flaw in the `/create_project` endpoint of UpTrain versions 0.7.1 and prior. The endpoint passes user-supplied `checks` and `metadata` parameters directly to Python's `eval()` function without sanitization, enabling arbitrary code execution. An attacker must have access to UpTrain and a valid authentication method (such as the default static API key). The vulnerable code uses `checks = eval(checks[0])` and `metadata = eval(metadata)`, allowing an attacker to inject malicious Python code that executes in the context of the UpTrain host process. As of publication, no patch is available; the project appears to be unmaintained as of March 2025.
Affected products
- UpTrain UpTrain 0.7.1 and prior
Timeline
- 2026-08-17: disclosed: CVE-2025-27770 published on NVD
- 2024-09-05: other: Issue opened in repo with no response from maintainers
- 2025-03-10: other: Slack messages indicate project is no longer being maintained
- 2026-08-10: other: Pull request opened with fix in repo