Committee hears NCSL briefing on state trends in artificial intelligence policy

Members of the Vermont House Energy and Digital Infrastructure Committee heard a briefing May 15 from Chelsea Canada of the National Conference of State Legislatures on emerging state trends in artificial intelligence policy, including comprehensive laws passed in Colorado and Utah, rising numbers of deep-fake and election-related measures, and guidance states are issuing to oversee AI use in government.

The presentation framed the topic as timely: states have been active in the absence of major federal regulation, introducing hundreds of bills in 2024 and more than a thousand proposals in 2025, Canada told the committee. She said many enacted measures focus on consumer protections, transparency, government use, and targeted use cases such as elections and explicit-content deepfakes.

Canada summarized three headline trends that have produced enacted state laws: comprehensive AI statutes that target “high‑risk” uses; narrower laws aimed at specific applications (for example, election- and deep‑fake-related restrictions); and a growing set of rules, inventories and procurement guidance governing state agencies’ use of AI. She told the committee that Colorado’s SB205 and Utah’s Artificial Intelligence Policy Act were examples of comprehensive approaches enacted last year and this year that other states are watching.

Colorado’s SB205, Canada said, applies to “high‑risk” systems and requires developers and deployers to take steps to avoid algorithmic discrimination and to perform risk management and impact assessments. The Colorado law, she said, grants the state attorney general enforcement authority, does not create a private right of action, and includes civil penalties of up to $20,000 per violation; the bill will not take effect until 2026.

Canada described Utah’s laws as imposing disclosure obligations for certain uses of generative AI and creating two new entities to study and advise on AI: an Office of Artificial Intelligence Policy and an AI Learning Laboratory. She said Utah’s follow-up measure this year (SB226) requires disclosures for generative AI and chatbots in some consumer and regulated services and ties violations to consumer‑protection enforcement.

The presentation noted a wide variety of other enacted and proposed measures across the states: bills requiring disclosure or watermarking for AI‑generated images, laws addressing deepfakes used in elections, statutes expanding child‑sexual‑image prohibitions to include AI‑generated images, protections for performers’ publicity rights, and targeted limits on some algorithmic pricing uses. Canada cited Colorado HB1004 (a measure that would limit use of pricing algorithms for residential rent-setting) and Arkansas’s recent clarification of ownership of AI‑generated content among state examples discussed in 2025 sessions.

Canada explained that states are also focusing on government use of AI. She said at least 10 states have instructed agencies to inventory automated decision systems; several states require impact assessments to check for bias, discrimination and disparate impact. She cited examples of state guidance that reference the National Institute of Standards and Technology’s AI Risk Management Framework (NIST) and noted that some states (including New York and California) have published procurement and use guidance that require human oversight, testing and a documented problem definition before procurement or deployment.

Vermont and Ohio were given as state examples: Vermont created an AI task force in 2018 and later established a Division of Artificial Intelligence within the Agency of Digital Services to inventory automated decision systems, draft a code of ethics for government AI use and establish an AI advisory council to advise the director. Ohio’s procurement policy for generative AI requires multi‑agency review, a risk assessment, privacy and security reviews, and limits generative training to public‑record data, Canada said.

Committee members asked technical and policy questions during the presentation. Representative Richard Bailey asked whether an “AI watermark” would automatically label AI‑generated content; Canada described industry proposals for visible and embedded provenance markers and noted that legislation varies, with some proposals requiring physical watermarks and others requiring data provenance records. Bailey asked: "Is this something that automatically goes on whatever this thing generates that tells us it's generated by AI?" Canada answered that proposals vary and industry and legislative approaches differ on visible watermarks versus provenance records.

Canada also described NCSL resources for legislators and staff, including a public AI policy toolkit, tracking of state bills since 2019, a series of briefs on AI and elections, workplace use, government use, and a 2024 survey of legislative staff that found staff use and interest in generative AI tools (chatbots, office productivity, coding assistance) and widely varying office policies on usage.

No formal committee action or vote followed the briefing; the session was a presentation and discussion. Committee members asked to follow up with NCSL for deeper information and Canada offered to provide additional materials and contact points.