AUD-918431 · 2026-07-26 11:24:57.508
Step-up enforced
Full reconstruction of one AI decision: the prompt, the retrieved evidence, the model output, every policy evaluated and the human who signed it. This record is hash-chained to its predecessor and cannot be altered without breaking the chain.
Risk score
71/100
at time of decision
Policies evaluated
2
inline at the gateway
Evidence documents
2
retrieved and cited
Human approval
Auto (step-up)
attributable sign-off
Decision reconstruction
Verbatim from the ledger
Prompt
Score card transaction of ₹18,40,000 to a new beneficiary.
Model output
Risk 0.91 — step-up authentication enforced.
Reasoning recorded
Velocity and device-change features dominate; beneficiary added 4 minutes before attempt.
Documents retrieved
- txn/CRD-99120-jul26.json
- models/graphnn-v4-card.json
Policy evaluation
Every rule that ran
Chain integrity
Tamper-evident
- Record hash
- 66aa31…0f92
- Previous hash
- f0c284…5531
- Anchored
- RFC-3161 timestamp · 11:45 IST
- Retention
- 10 years (RBI)
- Agent
- AGT-1007
- Initiating user
- svc-fraud-01
Chain verified — no gaps detected in this segment.