RI

Executive view

AI trust posture — Meridian Bank

Every agent, model and prompt in the estate is evaluated continuously against regulation, internal policy and responsible-AI thresholds. Figures refresh from the governance gateway every 60 seconds.

92AI Trust Score

Composite of compliance, responsible AI, security and audit readiness across 50 governed agents.

Compliance score

94

7 frameworks mapped · 35 open violations

Responsible AI score

89

10 dimensions · 2 below target

Security score

91

OWASP LLM Top 10 · 1 control failing

Audit readiness

96

Evidence chain verified 04:00 IST

AI agents governed

50

43 production · 7 pilot

+4 this quarter

Active AI models

18

6 vendors · 3 in-house

+2 approved

AI requests today

25,148

peak 3.1k/hr at 12:00

+8.4% vs 7d avg

Compliance violations

35

9 breach SLA at risk

-6 week over week

High-risk decisions

212

credit, AML, collections

+11 vs yesterday

Approvals pending

12

3 past 4-hour SLA

median 22 min

Failed policy checks

487

of 25,148 evaluations

1.9% failure rate

Critical risks open

8

2 awaiting exec sign-off

-1 this week

AI usage and enforcement — last 24 hours

25,148 requests · 312 blocked · 118 escalated to a human

Risk distribution

50 governed agents by inherent risk

Low · 26Medium · 14High · 7Critical · 3

Compliance trend

Rolling 6 months

Violations by category

35 open compliance violations

Policy violations timeline

Opened vs closed, weekly

Agent risk heatmap

Responsible-AI dimension score by business domain

Domainfairnesssecurityprivacyoversightrobustness
Lending
82
91
88
94
79
Financial crime
90
95
93
97
88
Customer service
86
84
81
76
83
Collections
71
82
86
68
74
Onboarding / KYC
88
89
79
90
85
Treasury & markets
93
87
92
88
91

Model performance

Accuracy, grounding and latency by serving model

gpt-4.1 (Azure)

9,840 requests / 24h

Accuracy

94%

Hallucination

1.8%

p95 latency

820 ms

claude-sonnet (Bedrock)

6,120 requests / 24h

Accuracy

92%

Hallucination

2.1%

p95 latency

910 ms

gemini-2.5 (Vertex)

4,310 requests / 24h

Accuracy

90%

Hallucination

2.6%

p95 latency

760 ms

llama-3.1-70b (in-VPC)

3,180 requests / 24h

Accuracy

87%

Hallucination

3.4%

p95 latency

640 ms

XGBoost credit v7

1,698 requests / 24h

Accuracy

96%

Hallucination

0%

p95 latency

40 ms