The September 1 AI certification is a signature. Put a measurement behind it.
Connecticut adopted the NAIC model AI bulletin as Bulletin MC-25, and it asks every Connecticut domestic insurer to certify annually, on or before September 1, either that it does not use AI systems or that its use is substantially consistent with the bulletin. That certification is your company's own statement. What an independent measurement adds is a dated record of how your model actually behaves, so the signature stands on evidence instead of adjectives.
What MC-25 asks, in plain terms.
The bulletin follows the NAIC model pattern: insurers using AI systems maintain a written program with governance and controls proportionate to risk. Connecticut added the calendar: an annual certification, due on or before September 1, signed by the insurer itself. Programs built from policy documents alone share a weakness — when someone asks how the model itself actually behaves, the file has adjectives where it needs numbers.
The honest note, up front: nothing in MC-25 requires independent testing. The certification is your own statement, and we will never claim a regulation requires this measurement. The case for it is simpler: a signature is easier to stand behind when there is a dated, independent record underneath it.
A signed, dated screening measurement, mapped to the text Connecticut adopted.
The LLM Validation Report works from the model's internal signals while your real question types run against a model your company hosts. It maps where the model fabricates, how often, and on which topics, down to the specific flagged answers your team can verify before anyone signs the certification. The instrument's published performance: 0.852 AUC on models it had never seen (fabricated-entity discrimination, length-controlled, leave-one-model-out), in the lab's method paper (DOI: 10.5281/zenodo.21365655).
Because MC-25 adopts the NAIC model text, the provision-by-provision crosswalk on this site maps the report directly to the language your certification references: what the report addresses, what it supports, and what stays honestly outside scope. In scope are models your company hosts or controls, including private Azure OpenAI and AWS Bedrock deployments and fine-tunes; vendor-attested SaaS AI is out of scope, and we will say so. The test never touches policyholder data.
Still piloting? The pre-deployment Model Selection Study measures your candidate models before you commit — the cheapest moment to measure is before the choice is made.
Fixed fees, written engagement, nothing success-based.
Fees are fixed and never success-based: the LLM Validation Report is $9,500, Continuous Assurance is $6,500 per quarter for one model, and the pre-deployment Model Selection Study runs $4,500–$7,500 by candidate count. Full details and the founding client program are on the pricing page. Every engagement runs under a written agreement before any work begins.
Six weeks is enough. Barely.
Delivery is five business days from materials-complete: your question types received and your endpoint reachable. An engagement started in July puts a signed, dated measurement in the governance file well before September 1. Write the lab with your model setup and the certification date you are working toward, and you will get a written plan the same business day.
Email the lab