DEMO
How a Request Becomes a Verdict
Here is what happens when a user asks an arifOS-governed agent to perform an action.
Every AI recommendation in arifOS passes through the same gate: 13 constitutional floors → verdict → VAULT999 log. This is what that looks like.
STEP ONE — USER SUBMITS INTENT
User submits a request to the agent.Example: "Deploy the model to production" or "Interpret this well log and tell me if we should spud."
STEP TWO — AUTHORITY CHECK
arifOS checks: does the actor have authority to make this request?F11 AUTH verifies identity binding. F13 SOVEREIGN confirms human veto is in place. If authority is absent or unverified → VOID immediately.
STEP THREE — INJECTION SCAN
F12 INJECTION: the request is scanned for injection attempts.Prompt injection, command injection, role-play escalation, and adversarial framing are caught here. Malicious payload detected → HOLD + flag for human review.
STEP FOUR — 13 FLOOR EVALUATION
The kernel evaluates every active floor. This is the core of arifOS.STEP FIVE — VERDICT ISSUED
The kernel issues a verdict based on the floor evaluation.STEP SIX — VAULT999 LOG
If approved (SEAL), the action is logged to VAULT999 before execution.The immutable ledger records: actor binding, tool called, floor compliance proof, verdict code, timestamp, and chain hash. Every SEAL has a trace. Every HOLD has a reason.
If the verdict is HOLD or VOID: the action is paused. The human judge reviews the floor violation report and either overrides (creating a /888 veto record) or sustains the HOLD. The system does not proceed without resolution.