AI evaluation tool & third-party oversight

Where AI insurance regulation is heading — the escalation track beyond the principle-based bulletin. All grounded in the Law360 article. GROUNDED · Law360

The comprehensive-model-law debate

The NAIC’s Big Data Working Group held six discussions in 2025 on whether to draft a comprehensive AI model law, and issued a Request for Information (RFI) in May 2025. The divide: GROUNDED · Law360

Consumer groups want uniform standards to prevent harm from AI-driven coverage/claim decisions made without human oversight.
Industry argues existing frameworks may suffice and warns against premature regulation without clear evidence of gaps.

The AI Systems Evaluation Tool

In July 2025, the Big Data Working Group introduced a draft AI Systems Evaluation Tool — a framework of questionnaires and checklists regulators use to assess how insurers deploy AI, focused on consumer risk and financial stability. Key features: GROUNDED · Law360

It supplements existing examinations (it’s an exam aid, not a new statute).
It targets transparency, accountability, and fairness, and standardizes assessment of AI governance and risk management.
Adoption is voluntary at the state level, but insurers are subject to it where a regulator chooses to apply it in an exam.
Pilot programs were expected in early 2026; pilot results would help decide whether a model law is needed. GROUNDED · Law360

Why this is gold for Glacis: the Evaluation Tool is essentially a published checklist of what regulators want to see. That checklist becomes a product requirements spec — build the platform so it answers the questionnaire by design, and compliance is a feature, not a scramble. GROUNDED · Glacis

Third-Party Data and Models Working Group (the vendor-licensing watch item)

This is the part that touches Glacis directly. AI vendors are currently not directly regulated by insurance departments, so the NAIC formed a Third-Party Data and Models (H) Working Group. In 2025 it: GROUNDED · Law360

adopted a broad definition of “third party” — any nongovernmental entity providing data, models, or outputs for insurance activities;
foreshadowed a model law on third-party oversight anticipated in 2026, potentially including LICENSING requirements for vendors;
told insurers to prepare for stricter diligence — contractual controls, documentation of model origins, and explainability standards.

Named example vendors in the article: Simplifai (workflow automation), Shift (fraud detection), Tractable (claims assessment). GROUNDED · Law360

Glacis is itself a “third party” under this broad definition — a provider of models/governance/outputs. That’s both a threat to manage (possible future licensing/explainability mandates) and a moat (Glacis already produces the documentation regulators will demand). See 02 The on-ramp / critical path. GROUNDED · Glacis

Disclosure, transparency & the global context

An emerging theme: should consumers be told when AI significantly affects a coverage/claims decision? The Evaluation Tool includes transparency questions; state regulators are exploring disclosure rules; and the EU AI Act will require transparency where AI is used in high-risk areas such as insurance. No formal US disclosure mandate exists yet, but it may appear in future model-law proposals. GROUNDED · Law360

State patchwork: Colorado leads

Some states wait for NAIC uniformity; others move first. Colorado’s Artificial Intelligence Act (passed May 2024) will require insurers to follow governance and testing procedures to prevent unfair discrimination — creating a patchwork that complicates multi-state compliance. GROUNDED · Law360

2026 outlook (insurer to-do list)

The article’s forward look: regulators likely begin using the Evaluation Tool in exams; a third-party model law is likely introduced; state initiatives increase. Insurers should audit AI systems, maintain detailed model inventories, and document testing — and demonstrate ongoing human oversight as agentic AI advances. GROUNDED · Law360

That to-do list is the Glacis feature list. See 00 Three revenue streams.

→ 02 AI claims litigation

Drill this page →5 bank questions stand behind what you just read. Check it while it’s warm.