Leading responsible AI use in your office

From personal use to team practice

As an officer in BPS 17 to 19 you are probably not only using AI yourself; people in your section are too, whether or not anyone has said so. Unspoken use is the riskiest kind: sensitive text pasted into public tools, unchecked drafts in files, no record of what was done. The aim of this final module is to turn scattered personal use into a simple, shared practice.

Start with what is allowed

Find out, in writing if possible:

• which AI tools your department has approved and for what kinds of information
• whether there is an organisational version (for example through Microsoft 365) with stronger data protection
• who to ask when something is unclear

If no guidance exists yet, raise it with your IT focal person or senior officer. In the meantime, apply the red lines from module 2.

Five rules for a section

A short, written set of rules works better than a long policy no one reads. For example:

1. Approved tools only for official work.
2. No classified material or personal data in public tools. Ever.
3. A person checks everything before it is signed, sent or filed, using the routine in module 7.
4. Say when AI helped with substantial documents, in the note to file.
5. Ask when unsure, and do not be punished for asking.

Keep a simple record

The National AI Policy 2025 calls for transparency about AI use in the public sector, including a public register of AI systems. You do not need a system for this yet, but a simple record helps: a shared list of the tasks your section uses AI for, which tools and who checks the output. When audits or registers arrive, you will be ready and the list will show where training is needed.

Know when not to use AI

Part of good judgement is choosing not to use a tool. Avoid AI where:

• the information cannot leave government systems and no approved tool exists
• a decision affects a citizen's rights or entitlements and no proper oversight process is in place
• trust, empathy or discretion matter more than speed
• you would not be able to explain how the output was produced

Help your team learn

• Share good prompts and templates that work for your section's routine tasks.
• Share mistakes too: a caught hallucination is a useful lesson for everyone.
• Point colleagues to this pathway and the how-to guides.
• Watch for over-reliance in junior staff, who may lack the experience to spot errors.

What comes next

The national policy envisages AI training for public servants, AI roadmaps for sectors including governance and guidelines from a national AI Directorate. As these arrive, update your section's rules to match. The habits from this pathway (protect information, keep a person accountable, check everything, be open) will remain the foundation whatever the tools become.

Try it

Draft five rules for AI use in your section, adapted from the list above to your work. Share them with one colleague and ask what is unclear.

Check yourself

1. Why is unspoken AI use in a team risky?
2. Name three of the five rules for a section.
3. Give two situations where you should choose not to use AI.

Answers: (1) Sensitive text may go into public tools, unchecked drafts may reach files and there is no record of what was done. (2) Any three of: approved tools only; no classified material or personal data in public tools; a person checks everything; say when AI helped; ask when unsure. (3) Any two of the situations listed under "Know when not to use AI".