AI Frontier Institute

Adds to existing law to establish provisions regarding unbiased artificial intelligence in state government purchasing.

View primary source →

Idaho House Bill 687 has passed the House (56-11-3) and is now in the Senate State Affairs Committee. It has not been enacted. If enacted, it would add Section 67-9236 to Idaho Code, covering how state agencies buy and use large language models (generative AI systems). Agencies, including higher-education institutions, could procure, deploy, or use such a model only if it is developed and implemented under two principles. The first is truth-seeking: the model prioritizes factual accuracy, historical fidelity, scientific inquiry and objectivity, and acknowledges uncertainty. The second is ideological neutrality: the model acts as a neutral, nonpartisan tool and does not intentionally manipulate outputs to advance or suppress an "ideological agenda," including DEI concepts. Ideological or policy-based responses are allowed only when the user expressly requests them or when they are clearly disclosed. Subsection (3)(a) bars agencies from using models designed, trained, fine-tuned, or configured to do any of four things. They are altering facts or historical representation to achieve DEI outcomes, suppressing or conditioning lawful content based on race or sex, embedding DEI constraints that override accuracy, or requiring users to adopt DEI concepts. Subsection (3)(b) still lets a model accurately describe DEI concepts, history, or critiques when a user asks. Under subsection (4), contracts must give the agency reasonable access, under confidentiality protections, to materials needed to verify compliance. Those materials include system prompts, model specifications, evaluation results, and source code or configuration files if needed. Contracts must also allow termination for material noncompliance after notice and a chance to cure, with the vendor bearing reasonable decommissioning or transition costs. Agencies may use state staff or independent experts for compliance reviews. Under subsection (5), each agency must adopt compliance procedures, including periodic performance review. It must consider whether distortions are systemic or intentional, the model's intended use, and the feasibility of corrective action. Subsection (6) says the section does not regulate private speech or non-state AI development and sale, does not mandate viewpoints or bar lawful academic inquiry, and creates no private right of action. Section 2 declares an emergency, so the act would take effect on passage and approval.

← Back to the tracker