House committee hears bill to limit state and local use of AI; sponsor open to law‑enforcement carve‑out

Representative Braxton Mitchell, sponsor of House Bill 178 and a member of the Montana House from House District 5 in the Flathead Valley, told the House Judiciary Committee the measure would limit how state and local governments use artificial intelligence, bar certain applications and require human oversight. "Any AI generated recommendation or decision that could affect a person's rights, privileges, or duties must be reviewed by a trained human officer," Mitchell said during the Oct. 12 hearing.

The bill as presented would: prohibit AI uses the sponsor described as behavioral manipulation, unlawful discrimination, malicious actions and inappropriate surveillance; require a disclosure when governments publish AI‑generated content that has not had human review; and mandate that trained human reviewers be able to modify or reject AI recommendations. Mitchell said he plans to work with the Department of Justice on a narrow law‑enforcement exemption.

Alex Sturhan, director of external affairs at the Montana Department of Justice, said the department rose in "soft opposition" and asked for changes to preserve investigatory tools in rural areas. "It can really help us in scalability and getting that out there," Sturhan said, noting examples such as bots that generate investigative leads for human agents in human‑trafficking work. Sturhan told legislators an exemption for law enforcement would be helpful and could change the department's position.

Informational witnesses described current government AI uses and limits. Kevin Gilbertson, Montana's state chief information officer, said the state is running a proof of concept in which cameras and AI assess road conditions and the AI then "update[s] the 5 0 1 site." Adam Carpenter, the state's chief data officer, explained the distinction between recommendation algorithms and autonomous decision systems and warned that definitions matter for how the law would apply to everyday software and to more independent systems.

Committee members pressed on multiple points: how the bill would treat traffic‑camera analytics and other Department of Transportation uses; whether forecasting models used by human services for Medicaid projections could be constrained; how the bill defines "cognitive behavioral manipulation" and whether the definition is broad enough to sweep in benign recommendation systems; and whether the bill should include penalties for misuse by government employees. Mitchell told members the cognitive‑manipulation language is an amendment the sponsor will circulate and that he is open to working with DOJ on a law‑enforcement carve‑out.

Witnesses and legislators also discussed bias and data quality. Carpenter and Gilbertson warned that AI systems can inherit biases from training data and stressed that safe use requires careful data practices and human review. Several committee members said they wanted clearer drafting so routine state uses — for example, traffic monitoring or website AB testing used to improve public outreach — would not be unintentionally barred.

No formal vote was taken during the hearing. The committee chair closed the hearing and scheduled executive action on bills, including HB 178, for the committee's next meeting. Mitchell said he is amenable to amendments and will work with stakeholders to refine statutory language.