SMSMC Blog
AI Security & GovernanceOctober 2026 · 7 min read

AI governance for towns, districts and utilities: borrow the federal and state playbooks

Local public bodies are not bound by federal or New York State AI policy, but those policies are free, tested templates. A NIST AI RMF-based first 90 days for councils, boards and their staff.

By Jim Venuto

Staff at a town hall, school district or water authority are already using AI: drafting letters, summarizing meetings, answering residents through a website chatbot, or using features that arrived inside software the organization already pays for. Most of those bodies have no AI policy, no inventory and no contract terms covering it. They also have something private firms do not: published, well-tested governance models from the federal government and New York State that they can borrow for free.

This post sets out what those models require, why they are a useful pattern even where they do not legally apply, and a practical first 90 days built on the NIST AI Risk Management Framework.

The gap is common

ISACA’s 2026 AI Pulse Poll found that only 38 percent of organizations have a formal, comprehensive AI policy, and 56 percent do not know how long it would take to halt an AI system after a security incident. The World Economic Forum’s Global Cybersecurity Outlook 2026 reports that the share of organizations assessing the security of AI tools before deployment rose from 37% in 2025 to 64% in 2026, but that roughly one-third still lack any such process. The same report found that 46% of small organizations report a lack of cybersecurity skills and expertise, compared with 29% of large ones. For state governments, NASCIO’s State CIO Top Ten for 2026 put artificial intelligence in the number one spot for the first time.

The framework: NIST AI RMF

The NIST AI Risk Management Framework 1.0 (January 2023) is voluntary and written to give “flexibility to organizations of all sizes.” It has four functions, GOVERN, MAP, MEASURE and MANAGE, and GOVERN applies across all of them. It describes trustworthy AI as valid and reliable, safe, secure and resilient, accountable and transparent, explainable and interpretable, privacy-enhanced, and fair with harmful bias managed. NIST’s companion Generative AI Profile (NIST AI 600-1) (July 2024) lists twelve risks for generative AI, including confabulation (“confidently stated but erroneous or false content”), data privacy, information integrity and information security. NIST’s AI RMF page notes that version 1.0 is being revised; nothing in a first-year program depends on the revision.

Two models a public body can borrow

Federal: OMB M-25-21 and M-25-22 (April 2025). These memos bind federal agencies, not local governments. M-25-21 requires each agency to designate a Chief AI Officer, to inventory its AI use cases at least annually and publish a public version, and to apply minimum practices to “high-impact” AI, meaning AI whose output is a principal basis for decisions with legal, material, binding or significant effect on people. Those practices include an AI impact assessment before deployment, testing even when the agency cannot see the vendor’s model or data, ongoing monitoring, human oversight, and a route for remedies or appeals. M-25-22 covers buying AI, including contracts that “permanently prohibit” vendors from using non-public agency inputs and outputs to train publicly or commercially available AI, and protections against vendor lock-in.

New York State: NYS-P24-001 and Appendix C-AI. The State’s Acceptable Use of Artificial Intelligence Technologies policy, last updated June 24, 2026, applies to state agencies; other governmental bodies are encouraged to consider it. Among its requirements: leadership approval before adopting new AI systems, an AI risk assessment for each system at least annually or on material change, a disclaimer on wholly AI-generated public content, and a flat rule that “automated final decision systems are not permitted.” It points agencies to the NIST AI RMF. The State’s Appendix C-AI standard terms for AI purchases is a ready-made contract model: AI tools not integral to a system disabled by default, written approval for AI use, and a clause that the vendor “shall not use Prompts, Uploaded Content, or Outputs to train, retrain, tune, or improve any AI model.”

New York’s LOADinG Act, as amended in February 2025 (S822, Chapter 96 of 2025), covers state agencies and now centers on a state AI inventory and published lists of automated employment decision tools. It does not reach towns, counties or school districts.

Local bodies are not bound by any of this. They should still use it: it is free, written by people who considered the same risks, and it is what auditors, residents and journalists will compare a local policy against.

A first 90 days

Days 1–30: GOVERN.

  • Name an accountable AI lead. It can be an existing official such as the IT director or clerk; M-25-21 allows federal agencies to use an existing official.
  • Adopt a short interim acceptable-use rule: no resident, student, patient or employee personal data in public AI tools, and nothing in a prompt that you would not release in response to a records request.
  • Require approval before any new AI tool or feature is switched on, and name an owner for each one.

Days 31–60: MAP.

  • Build the inventory: approved tools, free tools staff use without approval, and AI features embedded in existing software (office suites, records systems, help desk, cameras, plant systems). Record the data each touches and the decisions it informs.
  • Tier each use. Borrow the federal “high-impact” test and New York’s “directly impact the public” test: benefits, permits, hiring, student records, policing and public-safety support, and utility operations are the uses that need an impact review and a human making the final decision.

Days 61–90: MEASURE and MANAGE.

  • For high-impact uses, complete a short impact review: purpose, data, privacy and fairness risks, testing done, human oversight, how residents can challenge an outcome, and who accepts the remaining risk.
  • Write AI terms into new and renewing contracts, using Appendix C-AI and M-25-22 as models: notice before AI features are enabled, no training on your data, your ownership of outputs, an impact assessment or audit on request, data return, and a right to terminate.
  • Agree how to switch an AI system off after an incident, and test it once.
  • Brief the board or council on one page: what is in use, what was approved, the open risks, and the decisions only they can make.

For deployment security, use the joint guidance from CISA, NSA, FBI and partner agencies on Deploying AI Systems Securely (April 2024) and AI Data Security (May 2025). Counties can also draw on NACo’s AI County Compass, which includes sample policies and a procurement checklist.

Where SMC fits

SMC GRC.ai is Security Medic Consulting’s AI and quantum readiness practice. Jim Venuto founded SMC after a career leading security, privacy, GRC and CISO-level work for mid-size and large enterprises and public bodies, including AI governance. For a public body, SMC GRC.ai provides vendor-neutral leadership: the AI use policy and inventory, AI terms for procurement and contracts, privacy review where AI touches records about residents, students or patients, and briefings for boards, councils and audit committees. It does not sell AI products.

If your board or council is asking what AI is in use and who approved it, see AI and quantum readiness for public bodies at SMC GRC.ai. To talk it through, Book a free 30-minute call, or write to support@securitymedic.com.

Sources

Tags#nist-ai-rmf#sled#municipal#k12#playbook#executive