We take the security of trida-stack seriously. Thank you for helping keep the project and its users safe.

The latest release only. Older tags receive no fixes, no advisories and no SBOM.

This is a research project with a small maintainer group, and a support window we cannot honour is worse than a narrow one stated plainly. If you need a fix on an older tag, open an issue and say why -- we would rather have that conversation than imply a guarantee.

This line also defines a measured quantity. The supply-chain guideline computes SBOM coverage as releases with a valid SBOM / supported releases, so "supported" is the denominator: with the latest release carrying its SBOM, coverage is 100%, and the un-described older tags are out of scope by policy rather than by omission.

Please do not report security vulnerabilities through public GitHub issues, pull requests, or discussions.

Instead, report them privately by email to security@trillionlabs.co.

Please include as much of the following as you can:

• A description of the issue and its potential impact.
• Steps to reproduce, or a proof-of-concept.
• Affected files, commit, branch, or release.
• Any suggested remediation.

You may optionally encrypt sensitive details; ask in your first email and we will arrange a secure channel. Please give us a reasonable window to investigate and ship a fix before any public disclosure (coordinated disclosure).

• Acknowledgement: within 3 business days of your report.
• Initial assessment / triage: within 10 business days.
• Fix or mitigation plan: communicated after triage, with timelines that depend on severity and complexity.
• We will keep you informed of progress and let you know when a fix is released. With your permission, we are happy to credit you in the release notes.

In scope:

• Source code in this repository (the Apache-2.0 training and serving stack: train/, inference/, benchmark/ wrappers, and supporting scripts).
• The fetch/patch recipes and configuration shipped here.
• Build, packaging, and CI configuration in this repository.

Out of scope (report to the respective upstream project):

• Third-party runtime dependencies and backends installed separately — PyTorch, Hugging Face Transformers/Datasets/Accelerate, SGLang, vLLM, and the benchmark harnesses (see COMPLIANCE.md).
• The PolyForm-Noncommercial two-stream Gated-DeltaNet / block-diffusion kernels, which are not bundled here but fetched from upstream (yuchen-zhu-zyc/HybridDiffusion); report kernel issues upstream.
• Trida model weights and their inference behavior (model-card issues), unless the issue is in this repository's code.

This repository's git history was scrubbed of secrets prior to public release. If you nonetheless spot anything that looks like a credential, API key, token, private key, or other secret — in the current tree, the git history, CI logs, or an artifact — please report it privately to security@trillionlabs.co rather than opening a public issue or PR. Do not attempt to use any such credential.

Thank you for practicing responsible disclosure.