Threat Model (Dual-Agent Setup)

This document focuses on the dual-version model: ShadowNet Inside operates in a high-risk, hostile environment; ShadowNet Outside operates in a safer jurisdiction.

Assumptions

• Inside devices can be seized and inspected.
• Censors can monitor and interfere with network traffic (DPI, resets, blocking, throttling).
• Signal may be throttled or blocked periodically.
• Outside infrastructure and helpers are more protected, but still must assume opportunistic attackers.

Key goals

• Prevent config injection by untrusted parties.
• Minimize data on Inside devices (least persistence, encrypted at rest).
• Avoid outbound telemetry by default (Inside → Outside).
• Ensure connectivity recovery without user intervention when possible.

Threats and mitigations

• Device seizure (Inside)

- Encrypted local store: store.go - Minimal logging; avoid writing sensitive events to disk

• Malicious config injection

- Fail-closed signature verification: import.go - Trusted public keys configured out-of-band

• LLM misuse / data leakage

- LLM sees no secrets or full configs - Strict JSON-only bounded output and deterministic policy fallback

• Signal compromise / metadata risk

- Prefer disappearing messages - Keep message contents minimal (bundle URI only) - v2 encrypted bundles recommended for confidentiality

• Blackout scenarios

- Local Bluetooth mesh for peer-to-peer seed sharing