jNO follows a single-track release model — only the latest tagged release on PyPI receives security fixes. The main branch is the development line; security patches land there and ship in the next minor release.

If you believe you have found a security vulnerability in jNO, please do not file a public GitHub issue. Public disclosure before a fix is available puts other users at risk.

Instead, report privately by one of:

• Email the maintainer at leon.armbruster@iisb.fraunhofer.de with [jno security] in the subject line.
• Use GitHub's private vulnerability reporting (Security tab → "Report a vulnerability").

Please include:

1. A description of the vulnerability and what it allows.
2. The shortest reproduction you have (a single Python file, a domain construction, a malicious input file, etc.).
3. The jNO version (python -c "import jno; print(jno.__version__)") and any relevant JAX / hardware details.
4. Whether you would like to be credited and how.

The most security-relevant surface in jNO is:

• Loading checkpoints and pickled artefacts (jno.save, jno.load, statistics.load) — these use cloudpickle and execute arbitrary Python on load. Treat any checkpoint from an untrusted source as potentially malicious code.
• Loading meshes via meshio from user-supplied paths.
• The hyperparameter tuner running arbitrary user-supplied training configs.

Out of scope:

• Bugs that require already running arbitrary code in the same Python process as jNO (e.g., a malicious script importing jno and then calling its own functions). Those are application-level issues.
• Denial of service through excessive compute / memory in user-supplied PDEs (this is the user's responsibility — JAX gives them the gun, jNO just provides the holster).