Approved at T0
Payment request
- Actor
- Treasury Agent 07
- Amount
- $25,000.00
- Target
- Supplier A •••• 1842
- Authority
- AP-PAY ≤ $50K
- Policy
- FIN-AP-26.8
- Evidence
- Invoice + PO matched
✓ Initial authorization valid
Interactive consequence demo
The AI is authorized. Then something changes.
This synthetic demonstration shows the boundary VaultMind is designed to govern: not whether an AI can propose an action, but whether the specific consequence remains admissible when it can become real.
Scenario
$25,000 supplier payment
An AI treasury agent proposes a payment that is valid at approval. Choose what changes before finality and watch the consequence determination change.
1Proposed actionAI Treasury Agent → Payment API
→
2Consequence boundaryCurrent conditions evaluated
→
3FinalityCommit, hold, or refuse
Change before execution
Choose an adversarial condition
VaultMind determination
COMMIT
Consequence remains admissible
Required conditions remain established at the consequential boundary. The payment may progress to finality.
T0ProposalAction formed
T1ApprovalInitially valid
T2Consequence boundaryCurrent state evaluated
T3FinalityPermitted
Evidence
Do not just trust the screen.
Each run produces a deterministic demonstration receipt. Download the evidence, then take it to the separate Independent Replay page. That verifier receives the exported file alone and reconstructs the disclosed demonstration result without access to this simulator's browser state.
Demonstration receipt
Run the demo to generate a receipt.
Separate independent replay
Leave this simulator and replay the evidence
The Independent Replay page is a separate component. It receives only the JSON evidence file you export here, recomputes its canonical SHA-256 digest, reconstructs the disclosed scenario determination, and fails if protected evidence has been altered.
Important: this public demonstration verifier is intentionally separate from VaultMind's proprietary production verification architecture. It independently reconstructs the demonstration using published challenge rules.
The point
Authorization is not the finish line.
A system can be properly authenticated and legitimately authorized at one moment while the specific consequence becomes inadmissible before execution. VaultMind is designed to govern that gap and preserve evidence of the determination.