Vermont's Statutory AI Inventory — 16 Tools Disclosed, Only 2 Bias-Tested

Vermont's Act 132 (2022) created a Division of Artificial Intelligence inside the state's Agency of Digital Services (ADS) and requires ADS, under 3 V.S.A. Section 3303, to file an annual public report to the legislature that inventories the automated decision systems state government uses.
The January 2024 annual report identified 16 distinct AI tools in use across state agencies. ADS itself accounted for the most tools, mainly for cybersecurity threat detection. The Agency of Transportation uses AI to classify roadway pavement quality and to help prioritise sign-replacement and paving decisions. The Agency of Administration uses a tool called Gen TAX to screen tax filings for fraud risk. Agencies most often cited efficiency gains (6 tools), cost reduction (2) and improved security posture (2) as the tools' benefits.
The same inventory shows the limits of disclosure alone: of the 16 tools, only two — Bing Copilot and Azure OpenAI — had undergone any documented bias testing. Only two tools were flagged as directly affecting citizens: unspecified cybersecurity tools where AI makes final decisions on blocking intrusions, and an AI-assisted multi-factor-authentication system. An independent comparison by Government Technology (GovTech), published in October 2024, found neighbouring Connecticut's own inventory (10 tools, as of December 2023) was vaguer still — it listed general-purpose software such as Zoom and WordPress as "AI tools" and reported zero completed AI impact assessments, underscoring that Vermont's list, while incomplete on bias testing, set a comparatively higher bar for what counts as disclosure.
The honest lesson is that a statutory inventory is a necessary but not sufficient step toward accountability: naming the tools that exist does not by itself establish whether those tools are fair, accurate or safe to use.

Read the full analysis: https://digitalservices.vermont.gov/ai