ID H0687

State agencies may only procure or use large language models that prioritize factual accuracy, historical fidelity, and ideological neutrality, and must not intentionally manipulate outputs to advance political, social, or policy agendas including "diversity, equity, and inclusion" (DEI) concepts

Agencies are prohibited from using AI models designed to alter factual information for DEI purposes, suppress lawful content based on race or sex, embed DEI-based constraints that override accuracy, or require users to affirm DEI concepts to receive information

Vendor contracts must provide agencies access to system prompts, content policies, model specifications, training directives, evaluation methodologies, and source code to verify compliance, with contract termination permitted for material noncompliance

Each agency must adopt compliance procedures including periodic review of AI model performance, evaluating whether biases are systemic or intentional, the operational context, and technical feasibility of corrections

The law applies only to executive branch state government purchasing (including higher education institutions), does not regulate private AI development or sales, and takes effect immediately upon passage as emergency legislation