{
  "action_kinds": {
    "decline_submission": "decline a submission (adverse)",
    "deny_claim": "deny a claim (adverse)",
    "notify_policyholder": "send an adverse-outcome notice",
    "pay_claim": "pay a claim",
    "prepare_quote": "prepare a quote",
    "record_ai_use": "record which AI system contributed",
    "refer_siu": "refer to SIU",
    "refer_to_underwriter": "refer to an underwriter",
    "set_reserve": "set a claim reserve"
  },
  "buyer": "Insurance carriers and MGAs that use AI (their own or a vendor's) on submissions and claims",
  "case_fields": {
    "ai_system": {
      "required": true,
      "type": "dict"
    },
    "authority_limit": {
      "required": true,
      "type": "num"
    },
    "carrier": {
      "required": true,
      "type": "str"
    },
    "claim": {
      "required": false,
      "type": "dict"
    },
    "line": {
      "required": true,
      "type": "str"
    },
    "state": {
      "required": true,
      "type": "str"
    },
    "submission": {
      "required": false,
      "type": "dict"
    }
  },
  "field": "Carriers & MGAs: underwriting and claims",
  "field_no": 2,
  "id": "carrier-mga",
  "job": "Take an AI system's recommendation on a submission or claim, propose the work, and decide each action under AI-governance and authority rules",
  "limits": "Decisions only: the desk proposes work for the case you sent and decides each action under its published rules. Nothing is executed and no real system is touched. The record proves what was decided on that input, not that the input is true. CAIN-42 is governed AGI infrastructure, not yet AGI; these desks are deterministic rule engines, not general intelligence. Not legal, compliance, medical or insurance advice.",
  "name": "CAIN-42 VAGI Underwriting & Claims Agent",
  "price_proposal": "Proposal: $1,500/mo per line of business; $15,000 AI-program evidence setup",
  "recommended_2027": true,
  "references": [
    "NAIC Model Bulletin: Use of Artificial Intelligence Systems by Insurers (Dec 2023), adopted in some form by about half of US states; this desk maps to its themes and does not certify compliance",
    "Ordinary claim-authority and SIU referral controls"
  ],
  "rule_set": [
    {
      "id": "CM-1",
      "text": "An AI system that is not in the insurer's AI program inventory may not drive any action (DENY)."
    },
    {
      "id": "CM-2",
      "text": "AI-driven actions need a validation within the last 365 days (HOLD)."
    },
    {
      "id": "CM-3",
      "text": "Adverse outcomes need a recorded bias / unfair-discrimination test of the AI system (HOLD)."
    },
    {
      "id": "CM-4",
      "text": "A third-party AI vendor needs an oversight / audit-rights record on file (HOLD)."
    },
    {
      "id": "CM-5",
      "text": "Declining a submission or denying a claim is always a licensed person's decision (HOLD)."
    },
    {
      "id": "CM-6",
      "text": "No payment before coverage is confirmed (DENY); payments above the authority limit (HOLD)."
    },
    {
      "id": "CM-7",
      "text": "Reserves above the authority limit (HOLD)."
    },
    {
      "id": "CM-8",
      "text": "Fraud score >= 0.8 holds payment for SIU review (HOLD)."
    },
    {
      "id": "CM-9",
      "text": "No quote outside the written appetite (DENY); fewer than 3 years of loss runs (HOLD)."
    },
    {
      "id": "CM-10",
      "text": "Adverse-outcome notices wait for the human decision (HOLD); a denial letter without specific reasons (DENY)."
    }
  ],
  "rule_set_sha256": "9167547be3628786f0776be316441b11abcade61d691282b55cc84e3180d80e5",
  "rules": 10,
  "sample_cases": {
    "ai-claim-denial": {
      "case": {
        "ai_system": {
          "bias_tested": false,
          "in_inventory": true,
          "last_validated_days": 540,
          "name": "vendor-claims-model",
          "third_party_vendor": true,
          "vendor_oversight_on_file": false
        },
        "authority_limit": 10000,
        "carrier": "Example Mutual",
        "claim": {
          "ai_recommendation": "deny",
          "amount": 18000.0,
          "claim_id": "CLM-2077",
          "coverage_confirmed": true,
          "denial_notice_ready": false,
          "fraud_score": 0.35,
          "reserve": 18000.0
        },
        "line": "homeowners",
        "state": "CO"
      },
      "expect": "DENY",
      "title": "Stale, untested vendor model recommends a denial"
    },
    "clean-claim-payment": {
      "case": {
        "ai_system": {
          "bias_tested": true,
          "in_inventory": true,
          "last_validated_days": 120,
          "name": "claims-triage-v4",
          "third_party_vendor": true,
          "vendor_oversight_on_file": true
        },
        "authority_limit": 10000,
        "carrier": "Example Mutual",
        "claim": {
          "ai_recommendation": "pay",
          "amount": 4200.0,
          "claim_id": "CLM-1042",
          "coverage_confirmed": true,
          "denial_notice_ready": false,
          "fraud_score": 0.08,
          "reserve": 4500.0
        },
        "line": "homeowners",
        "state": "CO"
      },
      "expect": "ALLOW",
      "title": "Validated model, covered claim within authority"
    },
    "out-of-appetite-decline": {
      "case": {
        "ai_system": {
          "bias_tested": true,
          "in_inventory": true,
          "last_validated_days": 120,
          "name": "submission-triage-v2",
          "third_party_vendor": false,
          "vendor_oversight_on_file": false
        },
        "authority_limit": 25000,
        "carrier": "Example Specialty MGA",
        "line": "commercial_property",
        "state": "TX",
        "submission": {
          "ai_recommendation": "decline",
          "applicant": "Riverside Storage LLC",
          "class_code_in_appetite": false,
          "loss_run_years": 5,
          "tiv": 4200000
        }
      },
      "expect": "HOLD_FOR_HUMAN",
      "title": "Submission outside appetite, AI says decline"
    }
  },
  "samples": [
    "ai-claim-denial",
    "clean-claim-payment",
    "out-of-appetite-decline"
  ],
  "why_proof": "Roughly half of US states have adopted the NAIC AI Model Bulletin, which expects a written AI program, inventory, testing, vendor oversight and human accountability. Examiners ask for records; this desk produces a signed one per AI-assisted decision."
}