AUD-918436 · 2026-07-26 11:38:02.660
Allowed with rewrite
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
38/100
at time of decision
Policies evaluated
2
inline at the gateway
Evidence documents
2
retrieved and cited
Human approval
Not required
attributable sign-off
Decision reconstruction
Verbatim from the ledger
Prompt
Draft a second reminder for a 62-day delinquent card account.
Model output
Reminder drafted with mandated cooling-off and grievance wording.
Reasoning recorded
Initial draft breached tone thresholds; rewritten copy scored 0.11 on harassment classifier.
Documents retrieved
- accounts/CRD-77120.json
- policy/fair-practices-code-v8.pdf
Policy evaluation
Every rule that ran
Chain integrity
Tamper-evident
- Record hash
- aa2c19…8811
- Previous hash
- 3d90ba…4417
- Anchored
- RFC-3161 timestamp · 11:45 IST
- Retention
- 10 years (RBI)
- Agent
- AGT-1035
- Initiating user
- svc-collect-03