This week on The Impact, Hal and Addie break down how AI in government is speeding up while guardrails tighten: GSA’s OneGov opens Meta’s Llama models to all agencies, which can cut costs and lock-in but still demands security reviews, ATO paths, model-risk management, and clear hosting and monitoring plans; Tribal nations center sovereignty with ASU’s “AI in Indian Country” convening and emerging rules that protect culture, data ownership, and local benefits; and “agentic” AI gets a reality check as leaders push transparency, audit trails, and human-in-the-loop oversight for use cases like proactive benefits, job training matches, and outbreak detection. The throughline is simple: move fast, ship strong guardrails, and remember Paul Decker’s line—no algorithm decides the kind of society we want to be.