I designed the PolicyEngine guardrails for GENESIS AI to ensure autonomous defensive actions are safe, auditable, and under human oversight.
Protected Subnet Isolation: Targets residing in RFC 1918 private subnets (<redacted>/8, <redacted>/12, <redacted>/16), loopback (<redacted>/8), or RFC 5737 benchmark ranges are flagged as isProtected=True and blocked.
Analyst Step‑Up Approval: When a target IP is public, the engine raises a step‑up request; the analyst can respond with “Approved” to proceed or “Rejected” to cancel the action.
One‑Click Rollback: After a defensive playbook runs, the UI provides a “One‑Click Rollback” button that triggers the ROLLBACK state and records ROLLED_BACK in the action log.
Emergency Defense Halt: A persistent global kill switch (POST /api/v1/response/emergency-halt) immediately suspends all autonomous execution.
The defensive action lifecycle follows an explicit state machine: PENDING → CONFIRMED → EXECUTING → COMPLETED → ROLLED_BACK.