Do not open a public GitHub issue for security reports. Contact the maintainers privately through GitHub's private vulnerability reporting for this repository (the channel listed in .github/ISSUE_TEMPLATE/config.yml) with:

• affected component and version/commit
• a minimal reproduction or proof of concept
• impact assessment

You will receive an acknowledgment; we aim to triage within one week. We credit reporters by default unless anonymity is requested.

Security is a first-class architectural concern. The threat model lives in the specification (docs/spec/security.md); its current posture:

• No production execution. Real execution is opt-in only, confined to isolated Linux network namespaces, and never touches the host network (ADR 0012); v1.0 is not a production network controller.
• No automatic privileged operations. Privileged namespace operations require root and an explicit opt-in flag pair (--execute --yes-i-know), and refuse on non-Linux platforms (ADR 0012).
• The default and safest mode is read-only / simulated.
• State files are untrusted input: parsing is strict and total; malformed or corrupted state is rejected loudly (hash verification against content addresses), never partially loaded or silently repaired.

Identity, authentication, authorization, capability-based execution, cryptographic state identity, signed transitions, provenance, rollback safety, replay-attack resistance, isolation between simulation and execution, least privilege, secure defaults. Each will be specified and ADR-recorded before the corresponding stage of the roadmap.

In scope: the RAHN codebase, its specification, and its execution/simulation engines. Out of scope: the host OS, third-party dependencies (report those upstream and notify us), and the development infrastructure.