U.S. state legislatures are the place the motion is for putting guardrails round synthetic intelligence applied sciences, given the shortage of significant federal regulation. The resounding defeat in Congress of a proposed moratorium on state-level AI regulation means states are free to proceed filling the hole.
A number of states have already enacted laws round using AI. All 50 states have launched varied AI-related laws in 2025.
4 points of AI specifically stand out from a regulatory perspective: authorities use of AI, AI in well being care, facial recognition and generative AI.
Authorities use of AI
The oversight and accountable use of AI are particularly essential within the public sector. Predictive AI – AI that performs statistical evaluation to make forecasts – has reworked many governmental features, from figuring out social companies eligibility to creating suggestions on prison justice sentencing and parole.
However the widespread use of algorithmic decision-making might have main hidden prices. Potential algorithmic harms posed by AI techniques used for presidency companies embody racial and gender biases.
Recognizing the potential for algorithmic harms, state legislatures have launched payments targeted on public sector use of AI, with emphasis on transparency, client protections and recognizing dangers of AI deployment.
A number of states have required AI builders to reveal dangers posed by their techniques. The Colorado Synthetic Intelligence Act contains transparency and disclosure necessities for builders of AI techniques concerned in making consequential selections, in addition to for many who deploy them.
Montana’s new “Right to Compute” legislation units necessities that AI builders undertake danger administration frameworks – strategies for addressing safety and privateness within the improvement course of – for AI techniques concerned in essential infrastructure. Some states have established our bodies that present oversight and regulatory authority, reminiscent of these laid out in New York’s SB 8755 invoice.
AI in well being care
Within the first half of 2025, 34 states launched over 250 AI-related well being payments. The payments usually fall into 4 classes: disclosure necessities, client safety, insurers’ use of AI and clinicians’ use of AI.
Payments about transparency outline necessities for data that AI system builders and organizations that deploy the techniques disclose.
Client safety payments purpose to maintain AI techniques from unfairly discriminating towards some individuals, and be sure that customers of the techniques have a method to contest selections made utilizing the expertise.
Quite a few payments in state legislatures purpose to control using AI in well being care, together with medical units like this electrocardiogram recorder.
VCG by way of Getty Pictures
Payments overlaying insurers present oversight of the payers’ use of AI to make selections about well being care approvals and funds. And payments about scientific makes use of of AI regulate use of the expertise in diagnosing and treating sufferers.
Facial recognition and surveillance
Within the U.S., a long-standing authorized doctrine that applies to privateness safety points, together with facial surveillance, is to guard particular person autonomy towards interference from the federal government. On this context, facial recognition applied sciences pose vital privateness challenges in addition to dangers from potential biases.
Facial recognition software program, generally utilized in predictive policing and nationwide safety, has exhibited biases towards individuals of colour and consequently is usually thought-about a risk to civil liberties. A pathbreaking research by laptop scientists Pleasure Buolamwini and Timnit Gebru discovered that facial recognition software program poses vital challenges for Black individuals and different traditionally deprived minorities. Facial recognition software program was much less more likely to accurately determine darker faces.
Bias additionally creeps into the info used to coach these algorithms, for instance when the composition of groups that information the event of such facial recognition software program lack variety.
By the top of 2024, 15 states within the U.S. had enacted legal guidelines to restrict the potential harms from facial recognition. Some parts of state-level rules are necessities on distributors to publish bias take a look at stories and knowledge administration practices, in addition to the necessity for human overview in using these applied sciences.
Porcha Woodruff was wrongly arrested for a carjacking in 2023 primarily based on facial recognition expertise.
AP Photograph/Carlos Osorio
Generative AI and basis fashions
The widespread use of generative AI has additionally prompted considerations from lawmakers in lots of states. Utah’s Synthetic Intelligence Coverage Act requires people and organizations to obviously disclose once they’re utilizing generative AI techniques to work together with somebody when that particular person asks if AI is getting used, although the legislature subsequently narrowed the scope to interactions that would contain shelling out recommendation or accumulating delicate data.
Final yr, California handed AB 2013, a generative AI legislation that requires builders to publish data on their web sites concerning the knowledge used to coach their AI techniques, together with basis fashions. Basis fashions are any AI mannequin that’s educated on extraordinarily massive datasets and that may be tailored to a variety of duties with out further coaching.
Attempting to fill the hole
Within the absence of a complete federal legislative framework, states have tried to handle the hole by transferring ahead with their very own legislative efforts. Whereas such a patchwork of legal guidelines could complicate AI builders’ compliance efforts, I imagine that states can present essential and wanted oversight on privateness, civil rights and client protections.
In the meantime, the Trump administration introduced its AI Motion Plan on July 23, 2025. The plan says “The Federal government should not allow AI-related Federal funding to be directed toward states with burdensome AI regulations … ”
The transfer might hinder state efforts to control AI if states must weigh rules which may run afoul of the administration’s definition of burdensome towards wanted federal funding for AI.
