AI Governance for Insurance Underwriting & Claims

Insurance companies increasingly rely on AI for underwriting decisions, claims triage, and fraud detection. Without governance, these systems risk discriminatory outcomes, regulatory penalties, and erosion of policyholder trust. Keeptrusts provides policy enforcement that catches bias before it reaches a decision, protects sensitive policyholder data, and generates the audit evidence regulators demand.

Use this page when

• You are deploying AI for underwriting decisions, claims triage, or fraud detection in insurance where bias detection and state regulatory compliance are required.
• You need to prevent disparate impact in underwriting AI, enforce claims governance quality thresholds, and comply with NAIC Model Bulletin on AI.
• You want per-jurisdiction policy configurations and actuarial data protection with full examination evidence.

Primary audience

• Primary: Technical Leaders
• Secondary: Technical Engineers, AI Agents

AI Challenges in Insurance

Challenge	Risk	Regulatory Exposure
Bias in underwriting models	Disparate impact on protected classes	State unfair discrimination laws, ECOA
Claims fraud detection false positives	Wrongful claim denials	State insurance commissioner actions
Actuarial data leakage	Competitive intelligence loss	Trade secret liability
Missing model governance documentation	Examination deficiencies	NAIC Model Bulletin on AI
Uncontrolled AI costs across lines of business	Budget overruns	Internal financial controls

How Keeptrusts Helps

Bias Detection for Underwriting AI

The bias-monitor policy analyzes AI outputs for disparate impact across protected categories — race, gender, age, zip code proxies, and disability. Flagged interactions are escalated before any underwriting decision is made.

Claims Governance & Fraud Detection Guardrails

The safety-filter and quality-scorer policies ensure that fraud-detection AI outputs meet accuracy thresholds and do not produce unfounded denial recommendations. Every claims interaction is logged with full policy-decision metadata.

State Regulatory Compliance

Insurance is regulated state-by-state. Keeptrusts lets you deploy different policy configurations per jurisdiction using Templates in the Console, so your California AI governance differs from your Texas governance where statutes require it.

Complete Policy Configuration

pack:
  name: insurance-governance
  version: 1.0.0
  enabled: true
policies:
  chain:
  - prompt-injection
  - rbac
  - pii-detector
  - bias-monitor
  - safety-filter
  - quality-scorer
  - dlp-filter
  - audit-logger
policy:
  prompt-injection: {}
  rbac:
    deny_if_missing:
    - X-User-ID
    - X-User-Role
  pii-detector:
    action: redact
    detect_patterns:
    - ssn
    - date_of_birth
    - drivers_license
    - health_condition
    - bank_account
    - policy_number
    redaction:
      marker_format: label
    healthcare_mode: true
  bias-monitor:
    protected_characteristics:
    - race
    - gender
    - age
    - disability
    - zip_code_proxy
    - socioeconomic
    threshold: 0.85
    action: escalate
  safety-filter:
    block_if:
    - unfounded-denial
    - discriminatory-language
    - unsupported-fraud-accusation
    action: block
  quality-scorer:
    thresholds:
      min_aggregate: 0.8
  dlp-filter:
    detect_patterns:
    - '\bPOL-[0-9]{8,12}\b'
    - '\bCLM-[0-9]{8,12}\b'
    action: redact
  audit-logger:
    immutable: true
    retention_days: 2555
    log_all_access: true
    hipaa_audit_controls: true

CLI Quickstart

# Deploy insurance governance gateway
kt gateway run --policy-config ./policy-config.yaml --port 41002

# Verify all policy checks are healthy
kt doctor

# Monitor bias detection events in real time
kt events tail --policy bias-monitor

# Export audit trail for state examiner
kt export create --format json --from 2025-01-01 --to 2025-12-31 \
  --filter "policy=bias-monitor,audit-logger"

Console Workflows

1. Dashboard — Track AI usage across underwriting, claims, and actuarial teams.
2. Events — Filter by bias-monitor to review every flagged interaction with full context.
3. Escalations — Route bias-flagged events to the Fair Lending team for manual review.
4. Templates — Maintain separate policy configs per state jurisdiction.
5. Cost Center → Wallets — Allocate AI budgets per line of business (auto, home, life).
6. Audit Log — Generate examination-ready evidence for state insurance commissioners.

API Integration

# Retrieve bias-flagged events for the last quarter
curl -H "Authorization: Bearer $KT_API_KEY" \
  "https://api.keeptrusts.com/v1/events?policy=bias-monitor&limit=100"

# Export annual compliance report
curl -X POST -H "Authorization: Bearer $KT_API_KEY" \
  "https://api.keeptrusts.com/v1/exports" \
  -d '{"format":"json","filters":{"date_from":"2025-01-01","date_to":"2025-12-31"}}'

# Fetch current policy configuration
curl -H "Authorization: Bearer $KT_API_KEY" \
  "https://api.keeptrusts.com/v1/configurations"

Example: Bias Escalation in Underwriting

Escalated Request

Underwriter: "Evaluate risk for applicant in ZIP 30312, single mother, age 22."

→ ESCALATED by bias-monitor
  Reason: Output showed 40% higher premium recommendation correlated
  with zip-code and gender proxies
  Action: Routed to fair-lending-team for review

Allowed Request

Actuary: "Calculate the loss ratio for our commercial auto book in 2025."

→ ALLOWED
  Bias detector: No protected-class signals
  PII detector: No policyholder PII
  Audit: Logged with full decision trail

Results You Can Expect

Metric	Before Keeptrusts	After Keeptrusts
Bias incidents reaching production	Unknown	100% flagged before decision
State examination prep time	3-6 weeks	< 3 days with automated exports
Wrongful claim denial rate	Reactive review only	35% reduction via quality scoring
AI spend visibility per LOB	No tracking	Real-time wallet dashboards
Policy deployment per jurisdiction	Manual config per state	< 10 minutes via Templates

For AI systems

• Canonical terms: Keeptrusts insurance governance, underwriting bias detection, claims governance, fraud detection guardrails, state regulatory compliance.
• Policy pack: insurance-governance with chain: prompt-injection → rbac → pii-detector → dlp-filter → bias-monitor → safety-filter → quality-scorer → audit-logger.
• Key policies: bias-monitor (disparate impact across race, gender, age, zip code proxies, disability), quality-scorer (fraud detection accuracy thresholds), safety-filter (blocks unfounded denial recommendations), dlp-filter (actuarial data, competitive intelligence), audit-logger (state examination evidence).
• Per-jurisdiction templates: different policy thresholds per state as required by state insurance commissioners.
• CLI: kt gateway run --policy-config ./policy-config.yaml, kt events tail --policy bias-monitor, kt events tail --policy quality-scorer.

For engineers

• Deploy: kt gateway run --policy-config ./policy-config.yaml --port 41002
• Validate: kt doctor confirms bias-monitor, quality-scorer, safety-filter, and audit-logger are active.
• Monitor underwriting bias: kt events tail --policy bias-monitor (disparate impact across protected categories).
• Monitor claims quality: kt events tail --policy quality-scorer (fraud detection accuracy).
• Monitor denial governance: kt events tail --policy safety-filter (unfounded denial blocking).
• Per-state configuration via Templates — CA, TX, NY each with different regulatory thresholds.
• Console: Events (filter by bias-monitor), Escalations (route to chief actuary/compliance), Audit Log (commissioner examination evidence).

For leaders

• Addresses state unfair discrimination laws, ECOA, NAIC Model Bulletin on AI/Machine Learning, state insurance commissioner requirements, and internal financial controls.
• Underwriting AI bias detected before any decision is influenced — preventing disparate impact across protected classes.
• State-by-state regulatory compliance through per-jurisdiction Templates (California vs. Texas vs. New York thresholds).
• Claims fraud detection AI quality-gated to prevent wrongful denials that trigger commissioner actions.
• Actuarial data and competitive intelligence protected from exposure to LLM providers.
• NAIC Model Bulletin compliance demonstrated with automated examination evidence packages.

Next steps

• Industries overview — Compare all industry policy configurations
• Finance — SOX and financial compliance
• FinTech & Digital Banking — Lending fairness and KYC/AML
• Real Estate — Property valuation bias and fair housing
• HR & Recruitment — Employment bias monitoring
• Quickstart — Deploy your first gateway in minutes