The NAIC model bulletin makes carriers answerable for every AI system that touches a policyholder — including the vendor's. Optimus makes the whole stack something you can put in front of a market-conduct exam.
NAIC AI BulletinState DOI ExamsUnfair Trade PracticesGDPR
The NAIC bulletin, adopted across state insurance departments, holds the carrier responsible for third-party AI it relies on. The cleanest way to govern a model is to own the machine it runs on.
02
Claims files concentrate sensitive data
Health details, injury photos, financial hardship, litigation posture — a single claim file concentrates more sensitive data than most systems in the enterprise. It should be reasoned over where it lives.
03
Decisions that must survive discovery
Adverse decisions get challenged — by regulators and by plaintiffs' counsel. Full local logs of what the model saw and said turn AI from a liability theory into an evidentiary record.
The pain point — compliance
The bulletin made AI governance a filing requirement.
State regulators now expect a written AIS program: governance, inventory, testing for unfair discrimination, and controls over third-party systems. Cloud AI puts the hardest part — the system itself — outside your control environment. On-prem closes the loop.
The regimes in play
NAIC Model Bulletin
Written AI systems program, vendor accountability, and evidence of governance — enforced through state adoption.
Market Conduct Exams
State DOIs examine how decisions were made — including the systems that made them.
Unfair Trade Practices Acts
Outcomes tested for unfair discrimination; you need the logs to prove yours.
GDPR / State Privacy
Policyholder data rights follow the data — including into any model that processed it.
What cloud AI puts at stake
Claim files with health data leaving your BAA-less vendor chain
No decision logs when a DOI asks how the model reached an outcome
Vendor model changes that silently shift underwriting outcomes
Discovery requests you can't answer because the vendor won't
Optimus removes this exposure entirely: no external calls, no subprocessors, no egress — the full AI stack runs inside the controls you already operate.
Use cases
What one machine solves in insurance.
Every workload below runs on the appliance, against your documents and systems, with full logs — no internet connection required.
FNOL intake and claims triage
Summarize first notice of loss, check coverage against the policy, and route the claim — with the whole file staying in the claims system.
Loss summaries from calls, forms, and photos
Coverage checks cited to policy language
Severity routing with logged rationale
Claims TriageFNOL
Underwriting file assembly
Turn submissions, loss runs, and inspection reports into a risk narrative the underwriter can sign.
Submission summaries in your appetite framework
Loss-run analysis across carriers
Referral memos with document citations
UnderwritingRisk Narratives
SIU fraud investigation
Fuse claim history, medical records, and adjuster notes into investigation summaries — without shipping any of it to a third party.
Cross-claim pattern summaries
Timeline reconstruction with sources
Referral packages for law enforcement
SIUFraud
Policy and endorsement review
Draft and check policy language, endorsements, and filings against your forms library and state requirements.
Clause comparison across form versions
State filing checklists
Plain-language summaries for agents
Policy DraftingFilings
Market-conduct exam readiness
Every AI-touched decision carries its full local log — the record the bulletin expects you to produce.
Decision logs tied to claim and policy IDs
AIS program evidence on demand
Outcome testing runs kept on-box
Exam ReadinessDecision Logs
Actuarial and reserving support
Summarize triangles, draft reserve commentary, and search prior studies over data that never leaves the actuarial share.
Reserve movement narratives
Assumption documentation drafts
Prior-study retrieval with citations
ActuarialReserving
Every AI-assisted decision, documented on hardware you control.
Reserve a build slot, or talk to the team about a sovereign deployment for your organization.