'Fear and excitement' are driving AI discourse in government
December 3, 2025
After two years of study, New America released a wide-ranging report on how states are using and thinking about AI. The takeaway: generative AI is touching every part of public service, and leaders feel both urgency and risk.
As one policy expert put it, people are either excited or freaked out-and ready to act. Another noted that what's most encouraging isn't blind enthusiasm, but the quality of the conversation finally happening inside government.
Why this matters for state and local leaders
Generative AI introduces new capability and new exposure at the same time. Decisions you make now will shape standards for procurement, talent, data use, and public trust for years.
This week's top issues to watch
- Election integrity and data use: A proposal to repurpose DHS's Systematic Alien Verification for Entitlements (SAVE) for bulk voter checks has secretaries of state warning of "unacceptable risks," including searches of U.S. citizens by birth. Context on the program: USCIS SAVE.
- Federal preemption vs. state authority: A new push in Congress could bundle AI law preemption with child online safety legislation. NASCIO's executive director urged Congress to reject efforts that would sideline state AI laws and enforcement.
- Computer science leaders weigh in: Two former chairs at the Association for Computing Machinery argue Congress must preserve state authority in AI governance, noting that "AI's impact is happening now." More on ACM's policy work: ACM U.S. Technology Policy Committee.
New episodes of the Priorities Podcast post each Wednesday on major platforms, covering the latest for state and local IT and policy leaders.
What you can do this week
- Map use cases and risks: Inventory every AI and automation tool in use or in pilot. Classify by function, data sensitivity, model type, and mission impact. Assign owners.
- Lock in core guardrails: Update policies for procurement, human-in-the-loop review, records retention, model updates, and incident response. Require documented risk assessments for high-impact use.
- Pilot with constraints: Start small, measure results, and publish criteria for success and shutdown. Include privacy, security, bias, and accessibility checks upfront.
- Prepare for a patchwork: Track state statutes and federal guidance with your CIO, CISO, and AG. Build a compliance matrix and test it against a live project.
- Election-related systems: Do not repurpose datasets beyond their original scope without legal review, public notice, and independent validation. Document authority, audit trails, and redress paths-especially if citizen data is queried at scale.
- Upskill your teams: Train policy, procurement, legal, and program staff on AI risk, prompts, and evaluation. If you need a quick starting point, see role-based options here: AI courses by job.
Bottom line
Fear and excitement can be useful signals. Channel them into clear policy, measured pilots, and steady training. Move fast on the basics, and make every new AI deployment auditable, explainable, and defensible.
Your membership also unlocks:
