GPU VulnDB

Database/AI/ML frameworks & serving

UpTrain: authenticated remote code execution via the checks and metadata parameters on /new_run

CVE-2025-27772AI/ML frameworks & servingcurated

Impact

The /new_run endpoint is the third route in UpTrain 0.7.1 and earlier that turns the caller-supplied checks and metadata parameters into executed code, running as the UpTrain process inside its Docker container. An authenticated user gets whatever that container holds: evaluation datasets, model API credentials, and network position. Because the same defect appears on three separate endpoints, the practical control is who can log in at all, not which route is exposed. No patch existed as of publication.

Who can reach it

Any authenticated UpTrain user able to reach the dashboard backend over the network. No privileged role required.

What to do

No fix is available as of publication. Restrict who can reach and authenticate to the dashboard, assume every account grants shell in the container, and keep the deployment isolated from GPU-fleet and cloud credentials. Decommission it if the evaluation workflow is inactive.

References

This entry is curated: imported from vendor advisories with machine assistance, not yet individually verified. Confirm against your vendor's advisory before acting, and report anything wrong.