Report suspected vulnerabilities through GitHub private vulnerability reporting. Include the affected RhinoForge version, a minimal reproduction, impact, and any relevant Rhino Launch and operator-asset versions or hashes.

Do not report vulnerabilities in a public issue, discussion, or pull request. Do not attach restricted binaries, operator assets, model checkpoints, secrets, personal data, or private infrastructure details. Describe how an authorized maintainer can reproduce the issue instead.

Maintainers will acknowledge the private report, investigate the supported release scope, coordinate remediation and disclosure when appropriate, and publish an advisory for confirmed issues that affect a public release.

This policy covers RhinoForge source and release artifacts published by the HUIXI-AI organization. Separately distributed Rhino Launch packages, operator assets, model checkpoints, third-party dependencies, and components of the preconfigured board environment remain subject to their owners' security processes. A report that appears to belong elsewhere may be redirected without exposing its contents publicly.

Only versions identified as supported in a published release are eligible for security fixes. Source candidates and experimental model profiles may be used to reproduce an issue but do not create a support commitment.

Generic Hugging Face loaders and Pi0.5 do not execute remote model code. RhinoVLA runtime_factory values are executable installed integrations and must come from reviewed configuration. Hy-Embodied norm_stats.pkl files are also executable inputs: the required SHA-256 pins reviewed bytes but does not make an untrusted pickle safe.

Treat model checkpoints, tokenizers, processors, configuration, and pickle-backed statistics as trusted inputs. Review their source, revision, license, digest, and configuration before loading them.

RhinoForge declares torch==2.10.0. PyTorch security advisories apply when they affect that version and a reachable deployment code path; consult the upstream advisory and platform-provider guidance. RhinoForge does not claim to remediate vulnerabilities in PyTorch or another upstream dependency.

The v1.0.0 source does not call torch.jit.script or load .pt2 packages, so PYSEC-2025-194 and PYSEC-2026-139 are not reachable through RhinoForge's tracked code. Applications that use those interfaces remain responsible for the upstream risk. Reassess the dependency before adding either interface.

Privileged board-device access is intended only for trusted users on a single-tenant host. Do not expose it to untrusted code or mutually untrusted tenants.