Spaces

First-week checklist for an AI agent in a security or consular office

This checklist walks a defence or foreign affairs office through the first week of letting an AI agent help with chat, starting read-only and keeping every sensitive matter with staff.

Runs in your browser. Nothing you enter is sent anywhere or saved.

Before you start

Offices in this field rightly move slowly. The first week is not about answering everything. It is about proving, in a controlled way, that routine published questions get a good reply and that every sensitive message reaches the right person fast.

Work down the list in order. Do not skip the early steps, because they set the limits for everything after. Share the finished list with your security and legal colleagues and ask them to review the setup before any reply goes out without a person looking first.

Tick each one as you go. 0 of 9 done.

[ ] Write down the services and questions you want the agent to cover first.
[ ] Gather the published answers: hours, steps, documents, fees and contacts.
[ ] Write the out-of-bounds list: classified, operational, personnel and unreleased matters.
[ ] Name the team and the out-of-hours contact for every handover.
[ ] Run the agent in read-only mode and compare its suggested replies with your own.
[ ] Test with invented messages, including a distressed one and one with sensitive data.
[ ] Have security and legal staff review the setup and the reference.
[ ] Switch to drafts that a person checks before sending.
[ ] Review every conversation from the week with the team.

A sensible order for the week

Spend the first two days on the reference and the limits, not on the technology. Use the middle days for read-only testing with colleagues playing recruits, travellers and suppliers. Keep the last days for drafts and a review meeting.

Do not be tempted to widen the scope in week one. If staff want the agent to take on another service, add it after the review, with its own reference and its own handover rule.

Signs the week went well

You do not need figures to judge it. Look for these signals instead.

  • Routine answers match your published pages word for word where needed.
  • Every test of a distress message reached a person at once.
  • Staff understood each handover note without asking the sender again.

FAQ

Should the agent send replies on its own in week one?

No. Begin read-only, then use drafts that staff check. Move to sending on routine topics only after your team has reviewed real conversations and is comfortable.

Who should review the setup before launch?

Your security, legal and communications colleagues, together with the team that answers the public. Each sees a different risk, and the review is quick when the reference is short and clear.

How much information does the agent need on day one?

Only the published answers for the first few services, plus your limits and handover contacts. A short, accurate reference beats a long one with gaps in it.

More free tools for defence and foreign affairs offices

Read the guide: Answer public and partner questions in chat, and keep every sensitive matter with a person.