Citizen Portal
Sign In

Get Full Government Meeting Transcripts, Videos, & Alerts Forever!

Get email alerts on the Environment And Energy Artificial Intelligence Policy topic

No spam. Unsubscribe anytime.

Vermont committee reviews state AI statutes, ADS inventory and pending ‘deepfake’ disclosure bill

3209395 · May 7, 2025
AI-Generated Content: All content on this page was generated by AI to highlight key points from the meeting. For complete details and context, we recommend watching the full video. so we can fix them.

Summary

The House Energy and Digital Infrastructure Committee reviewed Vermont's existing AI statutes and reporting requirements, the Agency of Digital Services' automated decision-system inventory, and a set of pending bills that would regulate deepfakes, consumer-facing AI, high-risk systems and an AI registry.

MONTPELIER, Vt. — The Vermont House Committee on Energy and Digital Infrastructure spent its May 6 meeting reviewing state laws and proposed legislation related to artificial intelligence, focusing on how state government uses automated decision systems and several bills that would regulate private-sector uses of AI.

Rick Sagle of the Office of Legislative Council briefed the committee on the history and text of recent Vermont actions on AI, saying the legislature first moved on the topic in 2018 with what he identified as Act 137 and later codified a set of requirements and entities in 2022. "This statute only regulates the state's use of AI," Sagle said, describing requirements that the Agency of Digital Services (ADS) maintain an annual inventory of automated decision systems used by the executive branch.

The inventory that ADS must produce, Sagle explained, requires agencies to report vendor names and a description of systems' capabilities, whether systems have been tested for bias by an independent third party or are untested, the system's purpose and proposed use, data storage and sharing practices, and estimated acquisition and operating costs. ADS posts the inventory as part of its annual reporting, and Sagle showed the committee how to find the inventory and related documents on ADS's website.

Sagle described statutory changes that created a Division of Artificial Intelligence within ADS and an AI Advisory Council to advise the division. The division's duties include reviewing AI systems developed, employed or procured by state government; drafting a state code of ethics for AI in government; and reporting annually on the extent and impact of AI use in executive-branch agencies, including whether any systems implicate Vermonters' legal rights or privacy.

On state employee use of generative tools, Sagle said the division has published guidelines for content-generating AI intended for executive-branch use. He noted those guidelines are advisory: "It's just a guideline," he said, adding the legislature required the guidance but does not impose a mechanism in the statute to directly police day-to-day compliance.

Committee members asked about statutory sunsets and reporting lines. Sagle noted the AI advisory council statute contains a scheduled repeal date in 2027 unless the General Assembly acts to retain it, and he described ongoing administrative steps that send inventory and council reports to the committee.

The briefing moved to pending bills the presenter identified as active or dormant. S.23, which passed the Senate before crossover and was in House Government Operations at the time of the briefing, would require a mandatory disclosure label for deepfake media that depicts a person and is distributed within 90 days of an election. Sagle read the draft disclosure required in some versions of the bill: "This media has been created or intentionally manipulated by digital technology or artificial intelligence." The bill would not ban such content but would require a disclosure to run for the full duration of a video or be stamped on an image.

Sagle described three other bills that had not advanced to the same degree. H.340 would apply consumer-protection rules to automated decision systems used by businesses for "consequential decisions," such as educational assessments, employment, housing or lending, with the stated aim of avoiding algorithmic discrimination. H.341 would require safety standards and impact assessments for what the bill calls "inherently dangerous AI systems," including systems used in critical infrastructure, vehicles, medical devices, or to influence elections or voters. H.365 would establish a registry for AI systems and social media platforms, requiring operators to register, provide contact and privacy-policy information, disclose data-collection practices, and pay an annual fee (the bill text Sagle cited proposed a $100 annual fee for AI-system operators).

Sagle said remedies for violations in several bills would rely on existing consumer-protection enforcement, including potential actions by the attorney general under the Vermont Consumer Protection Act. He also pointed committee members to national resources, including a policy toolkit from the National Conference of State Legislatures, which the committee has requested to inform its work.

No formal votes or committee actions were taken during the briefing portion of the session. Committee members scheduled to hear later from ADS and the division said they expected to follow up on implementation details, independent bias testing practices, and the scope of disclosure and enforcement provisions in the bills.