City and government AI use policy template

A free AI use policy template for cities, towns, special districts and government agencies, to copy or download in Word.

Sections
  1. 1. Purpose
  2. 2. Scope
  3. 3. Definitions
  4. 4. Approved tools and the approval process
  5. 5. What staff may and may not enter into AI tools
  6. 6. Public records and retention
  7. 7. The record of AI use
  8. 8. Accuracy and human review
  9. 9. Disclosure of AI-drafted public content
  10. 10. Criminal justice information
  11. 11. Health information
  12. 12. Procurement and contracts
  13. 13. AI agents
  14. 14. Security and incidents
  15. 15. Training
  16. 16. Roles
  17. 17. Exceptions
  18. 18. Violations
  19. 19. Review cycle
  20. 20. Acknowledgment

[City Name] Policy on the Use of Generative Artificial Intelligence

Where this says "the city", put the name of your own town, district, authority or agency.

This policy is a starting point to adapt with the city's own attorney and is not legal advice.

1. Purpose

[City Name] uses technology to serve residents. Generative AI tools can help staff draft, summarize and analyze, and they also create risks for privacy, accuracy, security and public records. This policy sets plain rules so staff can use approved tools with confidence and know what is not allowed.

2. Scope

This policy applies to all city employees, interns, volunteers and contractors who use AI tools for city business, on any device and any account, whether city-owned or personal.

Council members and members of boards and commissions may adopt this policy by written notice to the [City Manager or Agency Head]. Those who do not opt in are encouraged to follow it, and their AI use on city business can still be a public record.

3. Definitions

Generative AI means software that produces text, images, audio, code or other content in response to a prompt. This includes chatbots, writing assistants, image generators and AI features built into other software.

AI agent means an AI tool that can take actions on its own, such as sending email, changing files or running commands, and not only answer a question.

Approved tool means a generative AI tool, at a specific version or plan, that has been approved under Section 4 and appears on the city's approved tools list.

Personal account means any account on an AI tool that is not issued and managed by the city.

Prompt means anything a person types, pastes, uploads or says to an AI tool. Output means what the tool returns.

Sensitive information means the categories listed in Section 5.

4. Approved tools and the approval process

Staff may use only approved tools for city business. The [Department] of Information Technology maintains the approved tools list and publishes it on [location].

A tool is approved only after review by three groups together: Information Technology, the [City Clerk or Records Officer], and the [City Attorney or Legal]. Where the tool may touch criminal justice or health information, the CJIS point of contact or privacy officer also joins the review.

The review asks at least these questions:

  • What data does the tool collect, and where is it stored?
  • Does the vendor use city prompts, files or outputs to train or improve its models?
  • Can the city retrieve, export and delete prompts and outputs?
  • Does the tool keep a log of use the city can reach, for the period the city's retention schedule sets?
  • Does the vendor's contract meet Section 12?
  • Which data categories is the tool approved for, and which is it not?

Approval is specific. A tool approved for drafting a meeting agenda is not automatically approved for case files. The approved tools list states the permitted uses for each tool.

New AI features that appear inside software the city already uses are treated as new tools until reviewed.

5. What staff may and may not enter into AI tools

Staff may enter information that is already public or that is intended for public release, such as published ordinances, public meeting notices and general questions.

Unless the approved tools list says otherwise for that tool, staff must not enter any of the following:

  • Personal information about residents, employees or applicants, including names with account numbers, Social Security numbers, driver license numbers, financial account details and dates of birth.
  • Criminal justice information, as described in Section 10.
  • Protected health information and other health records, as described in Section 11.
  • Records that are confidential, privileged or exempt from disclosure under [State public records law], including attorney-client communications, personnel and disciplinary files, investigation files and pending litigation materials.
  • Security details such as passwords, access keys, network diagrams, vulnerability reports, building security plans and incident response plans.
  • Procurement information that is protected before award, such as sealed bids and evaluation scores.
  • Any information a vendor or partner has given the city under a confidentiality agreement.

When in doubt, do not enter it, and ask your supervisor or the Information Technology help desk.

6. Public records and retention

Prompts and outputs about city business can be public records under [State public records law]. This is true whatever the device or account used. Using a personal account for city business does not remove the record from the city's obligations.

Staff must follow these rules:

  • Treat prompts and outputs that relate to city business as records, and keep them under the city's records retention schedule, [State retention law or records schedule].
  • Brainstorming, rough drafts and other transitory use may carry little or no retention requirement under the schedule. Output that is used in a decision, a final document, a policy or a response to a resident is kept for the period that applies to the final document.
  • Do not rely on an AI tool as the place where records are kept. Save records that must be retained to the city's records system.
  • Tell the [City Clerk or Records Officer] promptly when a records request, subpoena or litigation hold may reach AI prompts or outputs.
  • Do not delete prompts or outputs that are subject to a pending request or hold.

The [City Clerk or Records Officer] decides how requests for AI prompts and outputs are answered, with advice from the [City Attorney or Legal].

7. The record of AI use

The city keeps its own record of AI use on city-managed devices, separate from any AI vendor's chat history. The record shows at least who used which tool, when, what was sent and what came back, and what this policy decided, where a control applies it.

[Department] names the system that holds this record and who may open it. Opening a record is itself recorded. The record can be searched by person and date, so the city can answer a records request, an audit or a question from the [City Council].

[City Manager or Agency Head] reports to the [City Council] at least [annually] on how AI is used across the city, from this record.

8. Accuracy and human review

Generative AI can produce text that sounds right and is wrong, incomplete or biased. A person owns every output they use.

  • Any AI output used in a decision, a legal or financial document, or anything sent to the public must be read and checked by a qualified staff member before use.
  • Staff must verify facts, figures, names, citations and legal references against reliable sources. Do not assume an AI-supplied citation exists.
  • AI must not make final decisions about benefits, licenses, employment, enforcement or service eligibility. A person makes and signs those decisions.
  • Staff should watch for outputs that treat people differently based on protected characteristics, and must report concerns to their supervisor.

9. Disclosure of AI-drafted public content

When AI materially drafted content that the city publishes or sends to the public, the department must say so in a short note, for example: "This content was drafted with the help of an AI tool and reviewed by city staff."

Light edits such as spelling, grammar or formatting do not need a note. A resident who contacts the city must be told when they are talking to an automated system rather than a person, and must be able to reach a person.

10. Criminal justice information

Criminal justice information (CJI) is governed by the FBI CJIS Security Policy, currently version 6.1, and by the city's own CJIS agreements and rules. Those rules apply in full to any use of AI tools.

  • No CJI may be entered into any AI tool unless the [Police Chief or criminal justice agency head] and the agency's CJIS point of contact have approved that tool in writing for that use.
  • Approval follows the CJIS Security Policy rules for external systems, including a signed agreement with the vendor where those rules require one.
  • The approval must be recorded on the approved tools list.
  • Suspected exposure of CJI to an AI tool is reported under Section 14 and to the CJIS point of contact within the time the city's CJIS agreement requires.

11. Health information

Where the city or a city department is a covered entity or business associate under HIPAA, protected health information (PHI) may be entered into an AI tool only if all of the following are true:

  • The [Privacy Officer] has approved the tool for that use.
  • A business associate agreement is in place with the vendor.
  • The use is permitted by the city's HIPAA policies.

Free or consumer versions of AI tools are not approved for PHI unless the [Privacy Officer] says so in writing. Health information held by the city outside HIPAA is still sensitive information under Section 5.

12. Procurement and contracts

Contracts for AI tools, and for any software that includes AI features, must be reviewed by the [City Attorney or Legal] and must address at least the following:

  • City prompts, files and outputs remain city data and are not used to train or improve the vendor's models, or those of any third party, without the city's written consent.
  • The location where city data is stored and processed, and any subcontractors that handle it.
  • Retention, export and deletion of city data, including at the end of the contract.
  • Breach notice to the city within [number] hours.
  • Support for the city's public records obligations, including retrieval of prompts and outputs.
  • Where applicable, the CJIS and HIPAA terms described in Sections 10 and 11.

Departments must not accept click-through terms for an AI tool on behalf of the city.

13. AI agents

An AI agent may run only if the approved tools list names it and states what it may do. That list names the systems it may reach and the actions it may take, and any action that changes or deletes records, sends messages outside the city or spends money needs a person's approval first.

A named staff member owns each agent. Information Technology must be able to contain an agent, and each containment is recorded.

14. Security and incidents

  • Use city-issued accounts and city-managed devices for city business. Do not use personal accounts for city business.
  • Do not share city credentials with an AI tool or connect an AI tool to city systems without Information Technology approval.
  • Do not install AI browser extensions, plug-ins or agents that read city data unless they appear on the approved tools list.
  • Report any suspected misuse, exposure of sensitive information or unapproved tool use to [Information Technology Security Contact] immediately and no later than [number] hours after discovery. Reporting in good faith will not be treated as misconduct in itself.

The city's existing computer use and monitoring policies continue to apply to AI tools used on city systems.

15. Training

Staff must complete [City Name] AI training before they use an approved tool, and again [annually or every number months]. Training covers this policy, the data limits in Section 5, records duties, checking outputs, and how to report an incident. Supervisors receive additional guidance on reviewing AI-assisted work.

16. Roles

  • The [City Manager or Agency Head] owns this policy and answers to the [City Council] for it.
  • Department heads make sure their staff know this policy and approve department-level uses within it.
  • Information Technology maintains the approved tools list, reviews tools and handles security incidents.
  • The [City Clerk or Records Officer] advises on retention and answers records requests.
  • The [City Attorney or Legal] advises on contracts, exemptions and legal risk.
  • The [Privacy Officer] and CJIS point of contact advise on health and criminal justice information.
  • Every user is responsible for following this policy and for the outputs they use.

17. Exceptions

A department may request an exception in writing to the [Information Technology Director]. The request states the tool, the data involved, the reason and the safeguards. An exception requires the approval of the Information Technology Director, the [City Clerk or Records Officer] and the [City Attorney or Legal], is time-limited, and is recorded. No exception may permit use of CJI or PHI contrary to Sections 10 and 11 or to applicable law.

18. Violations

Violating this policy may lead to loss of access to AI tools and to discipline under [City personnel rules], up to and including termination. Contractors may face contract remedies.

19. Review cycle

This policy is reviewed at least every [6 or 12] months by Information Technology, the [City Clerk or Records Officer] and the [City Attorney or Legal], and sooner when law, guidance or the city's tools change. Effective date: [date]. Last reviewed: [date]. Policy owner: [Title].

20. Acknowledgment

I have read and understand the [City Name] Policy on the Use of Generative Artificial Intelligence, and I agree to follow it.

Name: [Employee name] Signature: [Signature] Department: [Department] Date: [date]

See Verillian govern AI on a real device.