Skip to content

Security: Y-Computer/recipes

Security

SECURITY.md

Security

Y Recipes documents experiments. A source-audited recipe is not automatically safe to expose on a customer network.

Known risks in the featured upstream experiment

  • Privileged containers.
  • Host networking.
  • InfiniBand device access.
  • Remote SSH orchestration.
  • Runtime remote-code trust.
  • Unauthenticated vLLM and ComfyUI endpoints bound to all interfaces.
  • Mutable container references and unpinned model artifacts.
  • Read-write host mounts.
  • System-wide page-cache flushing.
  • No durable readiness timeout, authorization layer or automatic rollback.

Do not expose the upstream configuration directly to the public internet.

Y OS deployment baseline

A supported profile should use:

  • Pinned model revisions, hashes and container digests.
  • Private management and model-serving networks.
  • Authenticated TLS at every user-facing endpoint.
  • Least-privilege containers and read-only mounts where possible.
  • Explicit readiness, shutdown and recovery gates.
  • Input-rights checks and output provenance for media workflows.
  • Structured audit logs that do not record private prompts by default.

Reporting

Do not open a public issue containing credentials, private prompts, customer data or unpublished vulnerability details. Contact security@y.computer with a minimal description and a safe return address.

There aren't any published security advisories