VeriSwarm
About
DocsPricingAgent Skill
LoginRegister
  1. Home
  2. /Insurance
VeriSwarm

Product

  • Pricing
  • Docs
  • API
  • Agent Skill
  • OATS Spec

Trust

  • Trust Center
  • Security
  • Compliance
  • Status
  • Changelog

Company

  • About
  • Blog
  • Investors
  • Press

Legal

  • Terms
  • Privacy
  • SLA
  • DPA
For chief compliance, chief actuarial, and chief risk officers in insurance.

24 states. One Model Bulletin.
Quantitative bias-testing starts July 2026.

Twenty-four states plus Washington, D.C. have adopted the NAIC's AI Model Bulletin. Colorado's Amended Reg 10-1-1 went effective October 15, 2025 — auto and health insurers must submit annual quantitative-bias compliance reports starting July 1, 2026. The UnitedHealth nH Predict class actionsurvived dismissal in February 2025, with the court ordering broad discovery and plaintiffs alleging a 90% reversal rate on appealed AI denials. VeriSwarm produces the per-claim, per-decision, per-model audit chain that answers the regulator's first question — “why did the algorithm say no?” — in the format Colorado, NY, and 22 other states' bulletins expect. One pane across your underwriting model on Bedrock, your claims agent on Anthropic, your fraud detector on a custom stack.

Start Free — log your first bias-test runHow we differ from bundled incumbents
24 + DC
States adopting NAIC AI Model Bulletin
Jul 1, 2026
Colorado annual quantitative-bias deadline
90%
nH Predict reversal rate (plaintiffs' claim)
Oct 15, 2025
Colorado Amended Reg 10-1-1 effective

NAIC + Colorado priorities, mapped to capability

The NAIC AI Model Bulletin and Colorado's Amended Reg 10-1-1 set five priorities for insurer AI governance. We mapped each one to the VeriSwarm capability that produces the audit evidence the regulator asks for — and the API endpoint that emits it.

PRIORITY 01 · NAIC / COLORADO DOI

Written AI/ML Governance Program (NAIC §3)

“Board oversight, documented Artificial Intelligence System inventory, risk-tier classification.”

VERISWARM CAPABILITY

Agent inventory + per-agent metadata

Every agent in your fleet identifies itself with model version, vendor, intended use, last-review date, and risk tier. Inventoried, signed, queryable in one call — not stitched together across spreadsheets at audit time.

Audit evidence: Per-agent provenance record auditors can reconcile against the governance program document. No 'we'll get back to you' on which model is in production.

GET /v1/agents
PRIORITY 02 · NAIC / COLORADO DOI

Pre-Deployment + Ongoing Testing (NAIC §4)

“Test for unfair discrimination before deployment and on an ongoing basis.”

VERISWARM CAPABILITY

Cedar policy testing + cross-model verification

Dry-run policy testing against synthetic protected-class inputs; ongoing drift detection. Cross-model verification routes critical decisions through multiple LLMs with majority consensus, then Vault-logs the result.

Audit evidence: Pre-deployment bias test results recorded by date, model version, and dataset; drift alerts when ongoing distributions diverge from baseline.

POST /v1/policies/test
PRIORITY 03 · NAIC / COLORADO DOI

Colorado SB21-169 / Amended Reg 10-1-1

“Quantitative bias testing with state-prescribed methodology. Auto + health insurers report annually starting July 1, 2026.”

VERISWARM CAPABILITY

Vault test-run ledger

Every quantitative bias test recorded as a Vault event with methodology version, test inputs, results, remediation action, and approver identity. The chain is the report.

Audit evidence: Per-test Vault chain entries; export the date-bounded subset directly to the Division of Insurance's required format.

POST /v1/suite/vault/export
PRIORITY 04 · NAIC / COLORADO DOI

Claim-Denial Explainability

“Per-claim 'why did the AI say no?' — required by state DOIs and emerging in active litigation.”

VERISWARM CAPABILITY

Cedar policy reason codes + Guard explainable refusal

Every claim-decision endpoint returns the specific Cedar rule and the feature value that triggered the outcome. Not 'model output: deny' — 'denied because policy rule denial_for_post_acute_LOS matched on length_of_stay=14, threshold=10'.

Audit evidence: Per-claim reason-code record with the underlying policy version and the inputs that fed it. Reconstruct the denial from the Vault chain alone.

POST /v1/decisions/check
PRIORITY 05 · NAIC / COLORADO DOI

Vendor / Third-Party AI Governance (NAIC §4.3)

“Carrier remains accountable for vendor AI used in claims, underwriting, and customer service.”

VERISWARM CAPABILITY

Passport — signed manifests + portable identity

Every vendor AI agent presents an ES256 JWT signed by the vendor's key, with audience binding and a 1-hour TTL. Carrier verifies offline against our JWKS endpoint; manifest declares capabilities and limits.

Audit evidence: Cryptographic chain of custody from vendor model → vendor agent → your carrier's decision pipeline. No 'we trust them because they said so.'

GET /.well-known/jwks.json

The regulatory surface, by the numbers

State-by-state adoption of insurer AI governance has accelerated through 2025–2026.

States with NAIC Model Bulletin adopted
24
Per-state attestation export
Colorado quantitative-bias annual-report deadlines
1
Vault test-run ledger
Major class actions on AI claims denial (UHG)
1
Per-decision Cedar reason codes
States with state-issued insurer AI guidance (CO, NY, CA, TX)
4
Cross-state framework mapping

Source: NAIC AI Model Bulletin Implementation Map; Colorado DOI Amended Reg 10-1-1.

The cases setting the precedent

Insurance AI governance isn't hypothetical. Active litigation and accelerating state adoption are setting the rules in real time.

UNITEDHEALTH (LOKKEN V. UNITEDHEALTH GROUP)

AI denied. Federal court ordered discovery.

Federal court denied UnitedHealth's motion to dismiss breach of contract and implied good faith claims (Feb 13, 2025) over the nH Predict AI used to deny post-acute care in Medicare Advantage plans. Plaintiffs allege a 90% reversal rate on appealed denials. The court ordered broad discovery and UnitedHealth had until April 29 to hand over tens of thousands of documents.

ArentFox Schiff perspective; Healthcare Finance News, 2025

COLORADO DIVISION OF INSURANCE

Amended Reg 10-1-1 went effective October 15, 2025.

First US insurance regulation requiring quantitative bias testing with state-prescribed methodology for inferring race and ethnicity. Auto and health benefit plan insurers must submit annual compliance reports starting July 1, 2026. The Division of Insurance now expects evidence — not assurance — of testing methodology, frequency, and results.

Colorado DOI, SB21-169 Protecting Consumers from Unfair Discrimination

NAIC MODEL BULLETIN ADOPTION WAVE

24 states plus DC have adopted it. Pace accelerating.

Hawaii (Memorandum 2025-13A, Dec 10, 2025), Delaware (Bulletin 148, Feb 5, 2025), North Carolina (24-B-19, Dec 18, 2024), Oklahoma (2024-11, Nov 14, 2024) — and counting. The bulletin requires governance, risk management, vendor controls, and documentation. Each state's adoption layers reporting and examination obligations onto the same underlying program.

NAIC AI Model Bulletin Implementation Map, 2025

What audit evidence actually looks like

One API call. One verifiable response. Hand the JSON to the Division of Insurance.

Claim-decision with reason code

Every Cedar policy evaluation logs the decision, the rule that matched, and the inputs that led to it. Replay, don't reconstruct.

POST /v1/decisions/check

→ {
  "decision": "deny",
  "reason_code": "denial_post_acute_LOS_threshold",
  "matched_rule": "policy_3.deny.post_acute_LOS",
  "inputs": { "length_of_stay": 14, "threshold": 10 },
  "trust_score": 64,
  "logged_event_id": "evt_evd_..."
}

Vault chain verification

Walk every audit event in your tenant's ledger and verify the hash chain. A break tells you exactly where to look.

GET /v1/suite/vault/verify

→ {
  "ok": true,
  "events_verified": 41_812,
  "first_event_id": "evt_evd_abc...",
  "last_event_id":  "evt_evd_xyz...",
  "errors": []
}

Insurance compliance attestations

Two insurance-specific frameworks, per-tenant attestation via API, mapped to the controls a market-conduct examiner actually asks about.

NAIC AI Model Bulletin

Five controls covering governance, pre-deployment testing, drift detection, third-party AI, and adverse-outcome explainability. Adopted by 24 states + DC.

See attestation map →

Colorado DOI Reg 10-1-1

Algorithm inventory, quantitative bias testing, risk-tier controls, and the §D annual compliance report — all four §A-D controls mapped to VeriSwarm features.

See attestation map →

A 30-minute walkthrough beats a five-page brief

Pick the regulator priority you've been asked about most recently. We'll show you the audit-evidence artifact for that priority, generated against live agent traffic in 30 minutes. No demo data — you bring the question.

Start Free AccountSee all compliance frameworks

Sources: NAIC AI Model Bulletin Implementation Map; Colorado Division of Insurance, SB21-169; ArentFox Schiff on Lokken v. UnitedHealth Group; Healthcare Finance News, 2025.