Guardrails for government workers

When you finish, you can protect citizen data and keep AI use within your agency's rules and the public's trust.

Citizen and official information

Never put residents' names, addresses, identification numbers, case files, health or benefit information, law-enforcement records, or non-public government documents into a consumer AI tool unless your agency has approved that tool for that data. Citizens often share information because they must, not because they choose to.

Privacy and public-records laws apply, such as data-protection laws like the GDPR in Europe and similar national or regional laws, and security classifications within government. Follow your agency's policy and your country's rules. If you are unsure whether a tool is approved, ask your IT or records office before using it.

Records, transparency and accountability

Many governments must keep records and respond to public-records or freedom-of-information requests. AI prompts, outputs and drafts may count as records in some places. Ask your records officer how to handle them.

Be able to explain how an AI-assisted document was produced. If a resident or auditor asks, you should be able to say what the tool did, what you checked and who approved it. Transparency builds trust.

Fairness, accuracy and bias

AI can produce errors that sound confident, reflect biases from its training data and treat languages, names or communities unequally. In public service, such errors can deny people services or damage trust. Test outputs with diverse examples and have staff who know the community review them.

Be careful with tools that score, rank or predict about people. Ask whether they have been tested for fairness, whether humans review results and whether affected people can appeal. If not, do not rely on them for decisions.

Disclosure and saying no

Follow your agency's rules on telling the public when AI helped prepare material or when they are speaking with an automated system. Some governments require this. Never present a draft as verified policy.

Say no when asked to use unapproved tools with citizen data, to publish unreviewed AI text or to let a tool make a decision about a person alone. A helpful line: “I can draft it with an approved tool and have it reviewed before release.”

Practice now

List the types of data your office handles and mark which should never go into an unapproved tool.
Ask your records or IT office how AI prompts and outputs should be handled, or write the question you would ask.

Use any AI assistant you have. Do not paste private, client or patient information into it.

Check yourself

1. A coworker pastes a resident's benefit case into a free chatbot. What is the main problem?

2. Why test AI output with diverse examples?

3. Why ask your records office about AI prompts and outputs?

Ask the AI tutor

Ask anything about this lesson, in any language. The tutor is an AI: it can be wrong, and it never needs your personal details.

Next lesson: Grow with AI: your 30-day plan

AI-drafted. Checked by an automated validator and a separate AI reviewer; a person has not read every lesson. Translations and tutor answers are AI-made. Found a mistake? Tell us. AI can be wrong. General information only, not professional advice. Not affiliated with any AI company named. Current University is not accredited.