Underwriting, claims handling, fraud detection, and distribution

The trust layer for AI agents in regulated industries.

The insurance sector has four sub-verticals, each with its own action vocabulary, human-approval gating, and regulator binding. Pick the sub-vertical that matches the calling system's role; the action classes, the gating boundary, and the regulator evidence pack follow from there. AGAP® governance records the agent's decision, gates issuance and settlement behind human approval, and signs every decision into an audit chain a regulator can verify offline.

Four calling-side shapes; one shared platform.

Underwriting

AI-assisted risk pricing and exclusion drafting. The agent prices and drafts; the underwriter issues. Vortalis gates policy issuance behind underwriter approval and records the pricing rationale on the audit chain.

Action classes

policy.pricedexclusion.drafted

Human-approval gating

policy.priced and exclusion.drafted are gated behind underwriter approval for policy issuance.

Primary regulators (maps to)

  • PRA Insurance Rulebook SS1/23 (AI guidance)
  • EIOPA AI guidance
  • FCA Consumer Duty (retail insurance)
  • GDPR Articles 22 and 9

Claims handling

Claim triage, fraud flagging, and settlement drafting. The agent triages and drafts; the claims handler settles. Settlement above the operator's threshold is gated behind a claims-handler approval.

Action classes

claim.triagedfraud.flaggedsettlement.drafted

Human-approval gating

settlement.drafted is gated behind claims-handler approval for settlement above a configurable threshold; claim.triaged and fraud.flagged are recorded for supervisory review.

Primary regulators (maps to)

  • FCA Insurance Conduct of Business sourcebook (ICOBS)
  • Lloyd's market bulletins on AI
  • PRA SS1/23

Fraud detection

Multi-policy correlation, network-pattern detection, and customer flagging. This sub-vertical cross-links to the financial-services fraud-monitoring pack where an operator runs both. Customer-impacting actions stay behind a human investigator.

Action classes

customer.flaggedcase.opened

Human-approval gating

customer.flagged and case.opened are recorded for investigator review; any customer-impacting action stays behind a human investigator. See the financial-services fraud-monitoring pack for the shared pattern.

Primary regulators (maps to)

  • Insurance Fraud Bureau guidance
  • FCA Financial Crime Guide
  • GDPR (cross-policy correlation)

Distribution and broker-facing AI

Quote generation and recommendation delivery for broker-facing flows. The agent generates and recommends; the broker reviews anything client-facing.

Action classes

quote.generatedrecommendation.delivered

Human-approval gating

recommendation.delivered is gated behind broker review for any client-facing recommendation; quote.generated is recorded for supervisory review.

Primary regulators (maps to)

  • FCA SM&CR
  • FCA Insurance Conduct of Business sourcebook (ICOBS)
  • Broker professional indemnity

Upstream services Vortalis adapts for this sector.

  • Salesforce Financial Services CloudShipping

    vortalis_proxy/services/salesforce.py

    Salesforce adapter as code. Counterparty certification with the operator's Salesforce org and the carrier systems behind it is the operator's responsibility.

What this sector pack does not do.

Vortalis does not underwrite or bind policy; it governs the AI agent's action.

The platform records what the agent priced, what the policy decided, and what the underwriter approved. The actual binding on the policy administration system and the actual payment instruction happen in the operator's existing infrastructure. Vortalis records the decision and the upstream outcome; it does not underwrite, bind, or pay.

Counterparty certification with each carrier system is the operator's responsibility.

Vortalis ships the Salesforce adapter as code, not as a counterparty-certified production integration. Production certification with each carrier system, claims platform, and payment service sits with the operator. The adapter provides a starting shape; conformance against each counterparty's certification programme is the operator's track.

Fraud-pattern detection accuracy depends on the policy and claim data the operator connects.

The platform governs the agent's fraud-detection actions and records them on the audit chain. The accuracy of any multi-policy correlation or network-pattern detection depends entirely on the completeness and quality of the policy and claim data the operator connects. Vortalis does not warrant detection accuracy; it records what the agent decided and why.

Reinsurance and treaty arrangements remain outside the platform.

The platform governs the primary-insurance agent's actions. Reinsurance placement, treaty arrangements, and the ceding calculations behind them are outside the platform's scope. Vortalis does not model treaty terms or govern reinsurance actions.

The general-purpose honest limits sit at /product/security-architecture; this list is specific to the insurance pack.

How operators put the pack to work.

These are illustrative deployments, not customer references. No named customer testimonials appear here, because the first external engagements are still landing. Each scenario describes a plausible shape of a Vortalis deployment in this sector.

Illustration 1

A UK insurer deploying an underwriting agent on a Salesforce Financial Services Cloud org, governing PRA SS1/23 and the FCA Consumer Duty through Vortalis. The agent prices against a tokenised view of applicant data, and policy issuance stays behind the underwriter's approval, producing a signed audit trail a supervisor can verify offline.

Illustration 2

A claims operation running a triage and settlement-drafting agent, governing the FCA ICOBS and Lloyd's market bulletins on AI through Vortalis. Settlements above the operator's threshold are gated behind a claims-handler approval, and every fraud.flagged action is recorded for supervisory review.

Illustration 3

A broker deploying a quote-and-recommendation agent for client-facing distribution, governing FCA SM&CR and broker professional-indemnity obligations through Vortalis. Every recommendation.delivered action stays behind broker review, and the attribution chain records which authorised person approved it.

Signed. Offline-verifiable.

A sample PRA SS1/23 and FCA Consumer Duty evidence pack, showing the structure of a signed, offline-verifiable pack. Signed. Offline-verifiable.

Download the sample insurance evidence pack (PDF)

Sample artefact. This is a placeholder evidence pack for layout and review; a real signed, offline-verifiable pack replaces it before launch.

Bring Vortalis to your insurance agents.

Pick the sub-vertical above. Read the integration brief if you would rather start with the engineering detail. Talk to us first if you would rather start with a conversation about your threat model.