Committee hears multiple bills on AI, algorithmic risk, UI testing and digital IDs

A committee member presented a package of bills and concepts on April 1 addressing artificial intelligence, automated decision systems and several related digital-privacy matters.

The first proposal described is a broad AI accountability bill that would target automated decision-making systems that make consequential decisions in areas such as employment, housing, credit, education and health care. The presenter said the draft focuses on transparency (identifying when a system made a decision and whether a human was involved), bias testing, whistleblower protections and accountability when discrimination or harms occur. The presenter noted other states have considered similar measures and cited Colorado and Virginia as recent points of comparison.

A second, narrower bill (referred to in the hearing as focused on "dangerous systems") would target high-risk automated systems that could cause substantial harm — for example, vehicle control systems, medical-device decisioning, critical infrastructure management, election-influence systems and biometric surveillance without consent. The bill would require safety assessments, testing, reporting and limits on particularly risky dual-use systems.

The presenter also described a bill to require robust testing, user-centered design review and public reporting for Vermont’s unemployment-insurance technology (the new UI system), calling for diverse stakeholder testing and stronger oversight (the Joint Information Technology Oversight Committee was identified as a possible oversight venue). Vermont Legal Aid requested attention to the UI system because of reported bugs and access problems in the current production system.

Another proposed bill discussed digital IDs and mobile-ID safeguards. The presenter described risks such as compulsory smartphone dependence, constant tracking if a mobile ID is used, remote revocation of identity credentials, and potential exclusion of people without smartphones. The draft would require safeguards including limiting data-sharing to only what is necessary, preserving a physical-ID option, prohibiting government or vendor tracking beyond verification needs and introducing a government “kill switch” for compromised systems.

Committee members raised questions about defining discrimination when algorithms are involved, the distinction between algorithmic inferences and human judgment, how to apportion liability, and whether federal laws (for example, education- and privacy-related federal statutes) already cover parts of this space. The presenter said the bills are meant to be complementary and to increase transparency, safety assessments and public accountability.

No committee votes or formal actions occurred on these proposals during the hearing; the presenter invited further stakeholder engagement and technical input from industry, civil-society groups and state agencies.