Skip to content

Security: leofmarciano/weightspace

SECURITY.md

Security

Reporting a vulnerability

Please don't open a public issue. Use GitHub private vulnerability reporting or email leonardo.marciano@toolzz.me. You should get an answer within a week.

Threat model

The Python sandbox that the LLM uses runs in your process. It is not a security boundary: code written by the LLM can do anything your Python process can do. Run untrusted models or LLMs in a container or VM, and don't keep secrets in the environment you don't want them to reach.

Loading .pt checkpoints uses torch.load, which can run arbitrary code for untrusted files. Prefer .safetensors.

There aren't any published security advisories