Public service rules that apply to AI

In this lesson you'll

• Explain why public agencies hold AI to a higher standard than most employers
• Name the public service duties that apply to every AI use
• Use ETA's ADAPT framework to decide whether a task is a good fit for AI

Welcome

Welcome to AI in Government and Public Service from the Enterprise Technology Association (ETA). Cities, counties, states, school districts, special districts, tribal governments and federal offices all run on the same thing: public trust. Residents can't choose another government when a service goes wrong. That makes AI a real opportunity for agencies that are short on staff and long on backlogs, and a real test of how carefully a public body uses new tools. This course assumes you've completed Level 1: Foundations of the ETA Applied AI Professional Certificate (free).

Why the bar is higher in government

A private company that makes a mistake with AI usually loses a customer. A public agency that makes the same mistake can deny someone benefits, delay a permit they need to stay in business, or publish something that isn't true under the government's name. Public agencies also work under rules most companies don't: open records laws, open meetings laws, due process, equal protection, language access and accessibility requirements, and procurement rules for every tool they buy.

Duties that apply to every AI use

• Accuracy. Anything the agency publishes or decides must be correct. The person who signs it owns it, including errors an AI introduced.
• Due process. When a decision affects someone's rights, benefits or property, people need notice, a reason they can understand and a way to appeal to a person.
• Equal treatment. Services must work for every resident, including people who speak other languages, have disabilities or don't use the internet.
• Transparency. Residents and oversight bodies can ask how a decision was made. Your agency should be able to answer.
• Records. Work you create in an AI tool for public business can be a public record. Lesson 2 covers this.
• Stewardship. Public money buys the tools, so procurement rules, security reviews and contract terms apply.

A worked example from the federal government

The White House Office of Management and Budget's memo M-25-21 (April 2025) sets AI rules for federal agencies. It names a category called high-impact AI: AI whose output is a principal basis for decisions or actions with a legal, material, binding or significant effect on rights or safety. High-impact uses need extra steps such as pre-deployment testing, an impact assessment, ongoing monitoring and human oversight. State and local agencies aren't bound by OMB memos, but the idea transfers well: the closer AI gets to a decision about a person, the more checking it needs. Many states have their own executive orders or policies; your agency's chief information officer or legal counsel can tell you which apply.

Workplace example: a county clerk's office

A county clerk's office wanted to use AI to answer the hundreds of emails it gets each week about marriage licenses, property records and passport appointments. The team started with ADAPT. Assess: routine questions with published answers were a good fit; anything about a specific person's record wasn't. Direct: they briefed the tool with the office's own published FAQ. Assure: a clerk read every draft before sending. Protect: no resident's personal information went into the tool. Tune: each week they added the questions it got wrong to the FAQ. Response time dropped from three days to one.

Try it

Pick three tasks your team does every week. In your approved AI tool, ask:

Goal: Help me sort these three tasks by how well they fit AI support in a public agency.
Context: I work in [department] at [type of government]. The tasks are [task 1], [task 2], [task 3].
Constraints: For each task, say who is affected if the output is wrong, whether it touches a decision about a specific person, and what a person must still check. Don't recommend AI make any decision about a resident's rights, benefits or property.
Format: A table with one row per task.
Done criteria: I can explain each rating to my supervisor in one sentence.

Check the ratings yourself. Disagree with at least one, and write down why.

Key takeaways

• Public agencies work under duties most employers don't: due process, open records, equal treatment and procurement rules.
• The closer AI gets to a decision about a person, the more testing, checking and human oversight it needs.
• ADAPT (Assess, Direct, Assure, Protect, Tune) is a quick way to decide if a task fits AI.

Sources

• OMB M-25-21: Accelerating Federal Use of AI through Innovation, Governance, and Public Trust (The White House) (opens in a new tab)
• AI Risk Management Framework (NIST) (opens in a new tab)