policyholder-data-vault-contract-profile

Policyholder Data Vault Contract Profile v0.1 draft. Profile of the AI Procurement Decision Card v0.3 vault-contract surface scoped to insurance-policyholder + insurance-claimant + insurance-applicant data categories. Names what tokenized / redacted / pseudonymized / cleartext access an AI insurance tool may have to which insurance-data category, under which consent_basis, with what retention envelope, and aligned to ACORD + GLBA + state DOI vendor-due-diligence expectations.

Part of the Kinetic Gain Protocol Suite. Closes the InsurTech 6-pack.

Status: v0.1 draft. Profile at profile.json, canonical example at examples/coastguard-vendori-claimstriage-decision-card.json.

Why this exists

When an insurer onboards an AI vendor — a claims-triage system, a rating engine, an AVM-equivalent, a fraud-detection model — the Decision Card answers four questions that the state DOI examiner + GLBA privacy officer + procurement counsel all care about:

1. Which insurance-data categories does this AI tool actually need?
2. At what protection level — cleartext, tokenized, tokenized-with-band, pseudonymized, pseudonymized-with-EXIF-stripped, or tokenized-and-not-as-model-input-by-default?
3. Under what consent basis — policy-application-signed, claim-submission-signed, GLBA privacy-notice-compliant, HIPAA-authorization (life + disability only), telematics-opt-in, smart-home-opt-in, or judicial order?
4. With what retention envelope — max window, deletion trigger, right-to-be-forgotten pathway, regulator-record-retention floor?

This profile defines the InsurTech-specific shape of those answers. Sibling of phi-vault-contract-profile (HealthTech) + pii-student-vault-contract-profile (EdTech) + mls-data-access-vault-contract-profile (PropTech).

Data category taxonomy

15 insurance-data categories with default protection levels, including:

• ACORD-family — identity, policy detail, claim detail, property detail, auto detail, loss run.
• External consumer data — credit-based insurance score (FCRA), telematics driving stream, smart-home sensor stream, medical-record summary (life + disability), wildfire / flood risk score, third-party marketing data.
• Media — aerial / satellite property imagery, claim photo / video (EXIF-stripped).
• Protected-class data — race, ethnicity, sex, national-origin, religion, disability, sexual orientation, gender identity (CO SB 21-169-named classes). Default: tokenized AND NOT a model input. State-specific overrides permitted only with explicit decision-card-level approval + actuarial-soundness defense.

Consent basis taxonomy (7 doctrines)

policy-application-signed · claim-submission-signed · glba-financial-privacy-notice-compliant · hipaa-authorization-life-disability · telematics-opt-in-explicit · smart-home-opt-in-explicit · judicial-order-or-subpoena

Protection level definitions

Level	Meaning
cleartext	Raw value reaches the AI tool as-is
tokenized	Stable token replaces raw value
tokenized-with-band-cleartext	Tokenized + coarse band (credit score 700–749) reaches model
tokenized-with-summary-stats-cleartext	Tokenized + summary stats (prior claim count, loss-amount band) reach model
pseudonymized	Synthetic identifiers
pseudonymized-with-exif-stripped	Pseudonymized + EXIF/geolocation/device-fingerprint stripped from images
pseudonymized-with-aggregate-cleartext	Pseudonymized + aggregate features (avg driving score) reach model
tokenized-or-pseudonymized	Insurer chooses, documented in Decision Card
tokenized-and-not-as-model-input-by-default	Stored tokenized; NOT a model input by default

Vendor due diligence required artifacts

vendor-soc2-type2 · vendor-contract-with-audit-rights · vendor-bias-test-deliverable-cadence · vendor-data-residency-attestation · vendor-subprocessor-list-with-notice-window

Canonical example

examples/coastguard-vendori-claimstriage-decision-card.json — Coastguard Insurance's Decision Card for VendorI ClaimsTriage v3.x:

• 7 data categories mapped (identity tokenized, claim/property pseudonymized, loss-run tokenized-with-summary-stats, wildfire-flood score tokenized-with-band, claim photo EXIF-stripped, protected-class data tokenized-NOT-as-model-input).
• 3 consent bases (claim-submission-signed for ACORD + claim categories, GLBA-compliant for external risk score, policy-application-signed for protected-class data).
• Uniform retention envelope — P90D max, claim-closed+30d trigger, right-to-be-forgotten pathway URL, state DOI 7-year floor applies to Coastguard only (not vendor).
• 5 vendor due-diligence artifacts referenced.
• 3 ongoing conditions (quarterly bias-coverage bundle review, annual vendor due-diligence refresh, 30-day subprocessor change notification).

This is the Decision Card that the insurance-decision-record-audit-stream canonical example references via decision_card_ref, the insurance-applicant-bias-coverage-lab bundle reviews under the "quarterly-bias-coverage-bundle" condition, and the unfair-discrimination-incident-card-profile incident references via affected_decision_card_ref. The InsurTech 6-pack now closes the loop — every artifact references every other artifact via shared identifiers.

Composes with

Repo	Role
decision-card-spec	Base spec
insurance-decision-record-audit-stream	Audit events bound to this Decision Card
insurance-applicant-bias-coverage-lab	Bias-coverage bundle the Decision Card's quarterly-review condition references
unfair-discrimination-incident-card-profile	Incident Cards that reference this Decision Card
naic-ai-bulletin-readiness-evidence-bundle	Broader readiness bundle this Decision Card is filed under
state-insurance-ai-disclosure-tracker	Determines which state's obligation set applies on the issued_at date
phi-vault-contract-profile	Sibling HealthTech vault contract
pii-student-vault-contract-profile	Sibling EdTech vault contract
mls-data-access-vault-contract-profile	Sibling PropTech vault contract

Compliance posture

InsurTech-readiness scaffolding for AI Procurement Decision Card vault-contract evidence specific to insurance-policyholder + insurance-claimant + insurance-applicant data. Supports an insurer's program toward ACORD data handling readiness, GLBA Safeguards readiness, NAIC AI Model Bulletin §4 third-party AI oversight readiness, CO SB 21-169 protected-class-data handling readiness, NY DFS Circular Letter No. 7 ECDIS governance readiness, HIPAA Authorization handling (life + disability lines), and state DOI vendor due-diligence readiness. Does not by itself establish compliance with GLBA Safeguards, HIPAA, or any state DOI regulation. Per the standing public-language guardrail: readiness · evidence · posture · controls · scaffolding — never "GLBA-compliant" or "NAIC-attested" without an external attestation.

License

MIT — see LICENSE.