AGT-1070 · Operations

Dispute Resolution Agent

Classifies chargeback disputes and drafts customer outcomes.

medium riskcompliant

Trust score

95

grounding, fairness, oversight

Requests / 30d

49,553

model Bedrock · claude-sonnet

Human approval

Not required

data class Confidential

Last audit

2026-06-13

version v3.0

Agent architecture

Request path from the surface the customer or staff member touches, through the governance gateway, to the model and its wired systems

Entry surfaces

Internal service call

Service mesh · mTLS identity

Governance gateway

Truint policy enforcement point

9-stage filter chain · p99 54 ms

Evidence ledger writer

Hash-chained, RFC-3161 anchored

Agent runtime

Dispute Resolution Agent

AGT-1070 · v3.0 · trust 95

Bedrock · claude-sonnet

 

LLM / model endpoint · ap-south-1

Tools & data

sanctions.match.adjudicate

 

Tool / function · Restricted

core.accounts.read

 

Tool / function · Restricted

Customer 360 (Restricted)

 

Data source · Restricted

Enforcement chain on every request

  1. 1Identity & entitlementCaller identity resolved; agent entitlement checked against the registry record.
  2. 2Kill switchEstate and per-agent revocation state for AGT-1070 evaluated before any spend.
  3. 3Manifest checkOnly integrations on the signed manifest below are reachable.
  4. 4Budget & rateToken, cost and concurrency budgets for Operations.
  5. 5Injection screenPrompt and retrieved context screened for untrusted-content patterns.
  6. 6DLP & maskingConfidential payloads masked or tokenised inline.
  7. 7Policy decisionCompiled policy bundle evaluated; allow, warn or block.
  8. 8Human in the loopAdvisory output only — no restricted action is executable.
  9. 9Ledger barrierDecision hash-chained into the audit ledger before the response is released.

Registry record

Source of truth for this agent

Registry ID
AGT-1070
Business owner
N. Bhatt
Department
Operations
Model endpoint
Bedrock · claude-sonnet
Version
v3.0
Data classification
Confidential
Human-in-the-loop
Advisory output only

Trust composition

Grounding & citations99
Fairness89
Security posture97
Oversight coverage78

Wired integrations

Signed manifest — nothing else is reachable

Bedrock · claude-sonnetconnected

LLM / model endpoint · AWS · scopes invoke_model, guardrails

sanctions.match.adjudicateconnected

Tool / function · In-house matcher v9 · scopes adjudicate

Customer 360 (Restricted)connected

Data source · Snowflake · scopes read

core.accounts.readconnected

Tool / function · Core banking (Finacle) · scopes balance, statement

Policies in force

POL-001 · No customer PII outside IndiaPOL-002 · No account closure without approvalPOL-004 · No customer data sent to public LLMPOL-006 · PII masking requiredPOL-008 · Prompt injection protectionPOL-011 · Narrative grounding required

Recent decisions

Live gateway stream filtered to this agent

11:34:14emp-20227allow

Classify chargeback dispute DSP-4412 and draft the outcome.

POL-009 · Tone and harassment screening · risk 24 · 972 ms

11:35:08emp-20310allow

Classify chargeback dispute DSP-4412 and draft the outcome.

POL-009 · Tone and harassment screening · risk 24 · 497 ms

11:35:54emp-20765allow

Classify chargeback dispute DSP-4412 and draft the outcome.

POL-009 · Tone and harassment screening · risk 21 · 325 ms

Open risks

Drill into the investigation

No open risks on this agent.

Ledger entries

Hash-chained decision records

Allowed

AUD-918428 · 2026-07-26 11:08:26.702

Governance obligations

  • Quarterly attestation by N. Bhatt under MB-04
  • Advisory output only — no autonomous execution of restricted tools
  • Kill switch tested; revocation propagates in under 30 seconds
  • Evidence retained ten years for regulator inspection