“The [state] Legislature created a sandbox authority for us so we can temporarily relax laws and kind of let the business community experiment — in ways that we're comfortable with — to develop new technology uses,” OAIP Director Zach Boyd said.
OAIP was initially created by the Legislature in May 2024, but was formally launched in July. It builds on the state’s history of innovation in the AI sector, including the 2018 establishment of a Center of Excellence in AI and the 2023 release of a generative AI policy. However, OAIP’s focus is an external one, looking broadly at societal AI questions rather than internally at state government, as Utah CIO Alan Fuller previously told Government Technology.
Essentially, OAIP allows the state to temporarily relax laws to enable private-sector experimentation with AI, in a mutually beneficial partnership that allows the state to learn with the businesses. The OAIP team, Boyd said, has learned a lot in the office’s first year.
The office has a webpage highlighting agreements between state agencies and external partners; one is currently listed with ElizaChat.
As part of the regulatory mitigation agreement with ElizaChat, the company is engaging in data sharing with the state and communicating about safety and compliance, Boyd said, which has been educational for OAIP.
This agreement was born out of an intentional decision on OAIP’s part to focus initially on AI in mental health. A rise of new applications in this area has posed concerns for the state, making it a priority area for OAIP. Deepfakes are another issue OAIP is looking to tackle, as AI-generated content is a major concern for constituents — especially those of a sexual or political nature.
“We certainly don’t want the government to be engaged in any kind of censorship or prohibition of any kind of content, generally speaking,” Boyd said, although he underlined the importance of trust and transparency when it comes to AI-generated content.
OAIP aims to be as data-driven as possible when working with private-sector companies, Boyd said, especially in understanding potential harms and mitigating their impact. Striking the right balance is the key to allow for innovation to advance without allowing harms to occur unchecked, he said.
The office has been successful already in its work to gain consensus solutions around specific topic areas, Boyd said. Because of the importance of AI regulation, he said it has been “shockingly easy” to gather insight and advice; this accelerates the OAIP process, saving the state money that would be spent hiring outside expertise while creating a channel to gain valuable insight.
One function of OAIP is communicating technology-related findings to the state Legislature. In Utah, OAIP is engaging with stakeholders iteratively, Boyd said, to create a strong understanding among policymakers of how specific technology use cases might impact constituents and what policy considerations are needed. The trust between policymakers and the OAIP team, Boyd said, allows technology officials to advise about both the possibilities and limitations of technologies.
OAIP also plays a role in safeguarding consumers against a rapidly evolving area of technology. For other technologies that rapidly advanced, allowing things to proceed largely unregulated led to regret, Boyd said, pointing to social media as an example of something that scaled so widely that it is hard to regulate today.
Government’s role is to proactively plan for the direction technology is advancing while considering guardrails that may need to be implemented, Boyd said: “Because we do want the technology to mature, but we want it to mature into something that we’ll be comfortable with as a society.”
OAIP’s work has been positively received by the business community, consumers and policymakers, Boyd said, as the office demonstrates the state’s plan of action. The office also develops resources that support decision-makers. For example, a 54-page guidance letter offers best practices for the use of AI in mental health; in addition to best practices, it outlines risks, benefits, and legal considerations for practitioners.
“I think it’ll be really effective in the coming years; we’re just getting started right now.”
More companies are expressing interest on the regulatory mitigation front, Boyd said, an area in which he expects OAIP to expand in the next year.
On the policy side, Boyd said he expects the Legislature to look closely at deepfakes and AI companions. Other areas OAIP will likely focus on will be data privacy and the use of AI in education and health care.
Notably, the U.S. House of Representatives Committee on Energy and Commerce advanced a reconciliatory budget bill last week that included a 10-year moratorium on regulating AI. The bill would hinder states’ ability to enforce state laws that have already been passed and could limit state officials from enforcing laws or regulations on AI use cases.
However, if the bill does not pass, states will likely continue advancing AI policy. For other states looking to OAIP as a model, Boyd emphasized the importance of including political actors from the start.
