Yes, we have testing we could hand over
No rule fires on this answer, for a model that influences a decision reaching a consumer. A reviewer moves on.
Free snapshot · question 9 of 9
Could you show a carrier that your model does not disadvantage one group of people more than another?
This is usually called disparate impact testing or fairness testing. It means you compared your model's outcomes across groups of people, for example approval rates or average price by race, sex or age, and that you wrote down what you found.
Why a carrier asks
What the answer decides.
This is the question an insurance regulator is most associated with, and the one a carrier is least able to wave through. They are not asking whether your model is fair. They are asking whether you can hand them evidence, because without it the gap appears in their examination file rather than yours.
What the engine does with each answer
Derived, not written.
Each outcome below was produced by running Vaulith's rules engine on that answer, for a model that influences a decision reaching a consumer. The same rules run in the paid assessment, on your whole estate, with the arithmetic shown.
We have looked, but not in a sendable form
Will escalateR-BIAS-UNRECORDED
Fairness testing exists but there is no record a carrier could read, for the system you described
You have compared outcomes across groups of people, and nothing you could hand a reviewer shows it: no method, no threshold, no date, no result. To a carrier that is the same file as no test, with one difference that works against you: a reviewer who hears "we have looked" and is then handed nothing concludes that the looking found something. This is the question an insurance regulator is most associated with, so the carrier cannot note the absence and move on.
No, we have not looked at this
Will escalateR-NO-FAIRNESS-TESTING
No fairness testing on record for the system you described
Nothing shows whether the model produces different outcomes for different groups of people. This is the single question an insurance regulator is most associated with, so a carrier cannot simply note the absence and move on: they have to decide whether to accept it, and that decision gets made above the person reading your answers.
The examination items it touches
Where this lands in the questionnaire.
- AIS-5.1 Describe testing performed to detect unfair discrimination in AI system outputs, including methodology, protected classes considered, and cadence.NAIC AI model bulletin, Section 3 ¶1.1 and ¶3.2 (bias analysis) · AI Risk Evaluation Supplement v5.0, Exhibit C
- AIS-5.3 Provide the most recent testing results for each consumer-impacting system.NAIC AI model bulletin, Section 4 ¶2.4 · AI Risk Evaluation Supplement v5.0, Exhibit C
- AIS-2.3 Describe the risk classification applied to each AI system and the criteria that place a system in the highest tier.NAIC AI model bulletin, Section 3 ¶1.4 · AI Risk Evaluation Supplement v5.0, Exhibit B 3k
Answer it about your own estate.
The snapshot asks this and eight more, in about ninety seconds, with nothing leaving your browser, and returns every finding with its reasoning and its remediation.