Please do not report security vulnerabilities in public GitHub issues.

If you discover a security bug in quantpreflight (for example, an unbounded allocation when parsing malformed Safetensors or GGUF headers, path traversal when resolving shard filenames, or credential leakage on cross-origin HTTP redirects), report it privately through GitHub Security Advisories at:

https://github.com/layer8labs8/quantpreflight/security/advisories/new

Include a minimal reproduction (such as a synthetic header fixture) and the affected version. We aim to acknowledge reports within 72 hours.

• quantpreflight never unpickles .bin or .pt PyTorch archives and never executes checkpoint code (trust_remote_code is never evaluated).
• Safetensors and GGUF binary parsers enforce strict header byte ceilings (64 MiB max Safetensors JSON header, bounded GGUF metadata counts and string lengths) before allocating memory.
• Shard paths inside model.safetensors.index.json are confined to the checkpoint root; .. traversal, absolute paths, and URL-scheme overrides inside shard entries are rejected.
• HF_TOKEN / Authorization headers are stripped automatically whenever an HTTPS 302/307 redirect crosses to a different host (such as CDN blob storage).