Skip to content

Security: outerreaches/mzg2

Security

SECURITY.md

Security policy

MZG2 publishes a research codec and a normative version-2 format specification. The CPU reference codec, store reader, synthetic vectors, K3 adapter and HIP decoder are on main. The extracted HIP decoder has not yet passed its GPU execution gate; no production-hardened or hostile-input-supported version is claimed. The existing v0.1.0-draft release contains the earlier documents.

Trust model: the decoder assumes trusted local model stores. Decoder errors invalidate the entire destination buffer, including bytes written before the error was reported, and the tile checksum detects accidental corruption only — it is not cryptographic and is not an authenticity mechanism. Do not decode stores from untrusted sources.

The full model — defenses, limits, and the remaining implementation hardening (fuzzing, malformed-input matrix, hostile-input boundary definition) — lives in docs/security.md.

Reporting

Use the repository's private vulnerability reporting route if available, or the contact route listed on the maintainer's profile. Please do not open public issues for memory-safety reports before a fix is available.

There aren't any published security advisories