Please report suspected security vulnerabilities privately, not through a public issue.
- Preferred: open a private security advisory through GitHub's "Report a vulnerability" button on this repository's Security tab (GitHub Private Vulnerability Reporting is enabled).
- The maintainer will acknowledge on a best-effort basis. This is free
software maintained by a single maintainer (Apache-2.0, no SLA — see
SUPPORT.mdwhere present); please allow reasonable time before any public disclosure, and we will coordinate a fix and credit.
This project is spec- and reference-validated, not platform-certified: it
is built and tested against the AOM IAMF v1.1.0 specification, the AOM
reference tools (iamf-tools, libiamf), and FFmpeg / GPAC MP4Box — not
against any streaming platform's private ingest pipeline. Findings about those
upstream projects, when this project's own tooling surfaces them, are
disclosed to their maintainers through their own security processes and are
tracked separately; they are not vulnerabilities in this code.
In scope for a report here:
- Crashes, hangs, unbounded memory growth, or other denial-of-service on malformed or hostile input reaching the measurement stack (WAV/PCM readers, the ADM/BW64 reader) or the plugin CLI surface.
- Any path that lets crafted input cause file writes, code execution, or reads outside the input under analysis.
Out of scope:
- A validator correctly reporting that a nonconformant file is nonconformant.
- Behavior that requires an already-trusted, attacker-controlled toolchain
binary on
PATH(the reference encoders/decoders run only as explicit, configured subprocesses).
Pre-1.0: only the latest released version on each line receives fixes. The
differential suite gates both measurement implementations (numpy reference ↔
sentinel-dsp kernel); hostile-input fuzzing covers the core's parser and
container walk (see iamf-sentinel's SECURITY.md); reference decoders and
EAR run only as explicit, configured subprocesses.