GuardARC runs with elevated privileges (it signals processes, tears down sockets, and installs nftables rules), so bugs in those paths are security-sensitive.

• The monitoring/alerting path needs only read access to local socket state (ss) and the ability to run the Slack helper.
• The enforcement path (SIGSTOP/SIGCONT, ss --kill, nft install/remove) needs root. Grant the narrowest privilege that works for your deployment.
• GuardARC never authenticates to or connects to a monitored server. It must never hold remote credentials.

• Signalling is strictly by PID, never pattern-based — the guard must never be able to kill itself or an unrelated process.
• The guard self-excludes its own PID from counting and enforcement.
• All destructive actions are scoped to a specific target peer and owning PID.

Report privately to the maintainers (Dana Research Group, Technion) rather than opening a public issue. Include the version, platform, and a minimal reproduction.