Exhibit D
The data section of the NAIC AI Systems Evaluation Tool: a checklist of data-element categories, asking which ones feed your models and where each came from.
Exhibit D of the NAIC AI Systems Evaluation Tool is titled “AI Systems Data Details.” It is not a free-text lineage narrative. The exhibit hands the insurer a fixed list of data-element categories and asks, for each one, whether the element is used in developing or training the company’s AI models, and whether it was sourced internally or bought from a named third-party vendor.
The categories are the substance of the exhibit. The list runs to twenty-five rows: twenty-four named categories, alphabetical from aerial imagery to weather and taking in education level, facial or body recognition, geo-demographics, medical data including biometrics, personal financial information, and the Version 4.0 addition of reasonable accommodations or policy modifications, with a twenty-fifth row left open for non-traditional elements the table does not name.
Nobody assembles that list by accident. It reads as a map of the variables most likely to stand in for a protected class, which is why insurers tend to treat Exhibit D as a proxy discrimination exercise even though the form never uses the phrase. Alongside the category, the exhibit asks what type of AI system uses the data and how the company uses it across its insurance operations.
Exhibit D is where “various internal systems” stops being an acceptable answer: the form has already named the elements, so the only remaining question is whether you can say where each one came from. Our Exhibit D data documentation playbook sets out what an acceptable answer looks like, element by element.