Skip to content

Security: arjhinety/OpenGrad

Security

SECURITY.md

Security

Reporting a vulnerability

Report privately to arjhine@experimentalmachines.org (the maintainer contact in the README). GitHub's private vulnerability reporting is not enabled on this repository, so do not open a public issue for a vulnerability. Include what is affected, how to reproduce it, and its impact. You can expect an acknowledgement within 7 days.

In scope: the Python package (src/opengrad), the scripts, the local annotation server (src/opengrad/annotation/server.py, integrations/annotate-ui), the MCP server (integrations/opengrad-mcp), the CI workflow, and any credential or personal data found in a tracked file, a release or the git history.

What not to post

Do not include credentials, private datasets, model access tokens or sensitive hardware logs in issues or pull requests.

Known, documented findings

A GitHub-token-format string that came with upstream NVIDIA When2Call data sits in one row of the frozen P-DET population and in one audit tarball. It is not an OpenGrad credential and was reported upstream; the bytes are kept because the file is pinned evidence (reports/ERRATA.md §22). CI's hygiene scan (scripts/repo/check_publication_hygiene.py) fails on any credential it does not already list.

There aren't any published security advisories