Claims intake agent
- Captures first notice of loss, structures the claim against the Policyholder and Claim objects, and attaches evidence — with claimant PII locked at the property level.
- Ontology permissions
AI agents for insurance
Claims intake, fraud triage, and payout — as governed actions.
The Northwind Claims agents run intake, fraud triage, and payout approval as approvable, compensatable actions. Any payout of $5,000 or more routes to a human approver before a single dollar moves — and the whole decision is sealed into a record you can defend.
Aligned with NAIC · GDPR · SOC 2 · ISO 27001 · EU AI Act
Action Fabric
Dry-run›
Refund $240
risk: low · evidence
Proposed›
Close case #4471
risk: med · evidence
Pending approval›
Payout $5,000
risk: high · evidence
Executed
Email customer
risk: low · evidence
actions ▸ dry-run → propose → approve → execute → compensate
The regulatory pressure
An AI that can pay a claim can also overpay a fraudulent one.
Claims handling sits between speed and exposure: unfair-claims-practices rules, model-governance expectations, and GDPR rights over claimant data all apply the moment an agent touches a case. An ungoverned bot approving payouts is an unbounded liability. Cortex makes the payout an approvable action with a hard human-approval threshold and seals every decision into a defensible record.
NAIC · model governanceUnfair claims practicesGDPR · claimant dataSOC 2ISO 27001EU AI Act
Regulatory pressure
NAIC · model governanceUnfair claims practicesGDPR · claimant dataSOC 2ISO 27001EU AI Act
Typed objects
Restrictedlocked
frameworks ▸ encoded as policy · gated · sealed
Use cases
From first notice of loss to a defensible payout.
Each agent maps the claim to typed ontology objects, runs under an oversight mode tuned to its risk, and threads every action through the Action Fabric.
Fraud triage agent
- Scores claims for fraud signals and routes suspicious cases for investigation rather than silent denial — every score explainable and logged.
- Observability
Payout approver agent
- Proposes payouts as actions; anything ≥ $5,000 pends for human approval, and any payout can be compensated (reversed) with a sealed reason.
- Action Fabric
Claimant communications
- Drafts status updates and decision letters grounded to the claim record, with disclosure language enforced by policy, not left to the model's phrasing.
- Policy-as-Code
Subrogation & recovery
- Identifies recovery opportunities under hard budget caps, so a large batch can't burn through its spend before a human reviews the exceptions.
- Cost governance
The payout is a gate, not a guess
Payout ≥ $5,000 → human approval. Always. By policy.
The payout threshold is a fail-closed action gate, not a guideline an agent can talk its way past. Below it, low-risk payouts execute and seal; at or above it, the action pends for a named approver — and every approval, denial, and reversal is signed into the ledger.
- Payouts ≥ $5,000 pend for human approval before any money moves
- Every payout is compensatable — reverse it with an audited reason
- Fraud scores are explainable and logged, never a silent denial
- Claimant PII and medical detail locked at the property level under GDPR
Action Fabric
Dry-run›
Refund $240
risk: low · evidence
Proposed›
Close case #4471
risk: med · evidence
Pending approval›
Payout $5,000
risk: high · evidence
Executed
Email customer
risk: low · evidence
actions ▸ dry-run → propose → approve → execute → compensate
Prove it — don't just claim it
The verdicts an adjuster and a regulator will see.
These are the literal Action Fabric responses — the threshold is enforced, not advisory.
runtime · verdicts
Payout $5,000HOLD · approval
Payout $240200 sealed
Medical field403 redacted
Reversal200 compensated
fail-closed ▸ challenge it and it denies, on the record
- 01
Model the domain
Map the work to typed ontology objects — Policyholder, Claim, Payout, Adjuster, Evidence — with restricted fields like ssn / medical_detail locked at the property level.
- 02
Encode the rules
Translate the obligations above into Policy-as-Code and Action Fabric approvals — the gates fail closed, returning a precise code, not a guess.
- 03
Prove the outcome
Every run lands in a hash-chained Trust Ledger with signed receipts and 10-hop provenance — a record you can hand the regulator.
$5,000Payout approval threshold
10-hopProvenance on every decision
5 modesOversight settings per workflow
100Live-verified run quality score
Works with the rest of the runtime
The controls behind insurance.
The same shared runtime powers every vertical — here are the capabilities and the Solution Pack this industry composes from.
Action Fabric
Payouts run as approvable, compensatable actions with a hard approval threshold.
Oversight modes
Tune autonomy per workflow — auto-execute the small, hold the large.
Ontology
Model Claim, Payout, and Policyholder; lock medical and PII fields.
Trust Ledger
Every approval, denial, and reversal sealed into a defensible record.
Claims Solution Pack
Deploy a signed claims pack — verify the signature before you import.
Trust Center
Map gated payouts and sealed records to NAIC and GDPR obligations.
Security & compliance
One review for the claims floor and the compliance office.
The action gates, RBAC, DLP, and detection & response are shared across claims, fraud, and recovery — so you map once, to the obligations insurers already report against. Aligned with, never claiming certification you don't hold.
NAIC · model governanceUnfair claims practicesGDPR · claimant dataSOC 2 · auditISO 27001 · integrityEU AI Act · logging
FAQ
AI agents for insurance — questions, answered.
Can an AI agent approve an insurance payout by itself?
Only below the threshold you set. Payouts run through the Action Fabric as proposable actions; anything at or above $5,000 pends for a named human approver before any money moves, and every payout is reversible with an audited reason.
How does Cortex handle claimant privacy?
Claimant PII and medical detail are locked at the ontology property level (deny-by-default once governed) and redacted before any object reaches the model, supporting GDPR data-minimization on every claim.
What stops the fraud-triage agent from silently denying claims?
Fraud scores are explainable and logged, and suspicious cases route to human investigation rather than auto-denial. The triage decision and its reasoning are sealed into the Trust Ledger.
Can a wrong payout be reversed?
Yes. Every executed payout is a compensatable action — it can be reversed with a mandatory audited reason, and the reversal is sealed into the ledger alongside the original.
More industries
One governed runtime, every regulated vertical.
The controls are the same — the obligations they satisfy differ. Explore another vertical, or see the full set.
Pay claims fast — and keep the proof you can defend.
See the claims approval board, the $5,000 payout gate, and the sealed record behind every decision.