Easy summary: Use approved tools, keep protected information out unless the use is authorised, check important outputs, and record concerns. If AI affects a decision about you, ask what was used, what information mattered and who can review the result.
Pause before you paste
| Information | Safer next step |
|---|---|
| Public, non-sensitive information | Check the tool is approved and still verify the output. |
| Internal business information | Check policy, tool settings, retention and who may receive the data. |
| Personal information | Do not enter it unless the purpose, authority and privacy controls are clear. |
| Health, identity, financial or other sensitive information | Stop and obtain specific approval; a public chatbot is rarely the right place. |
| Confidential, privileged or third-party material | Check confidentiality, contract, copyright and access restrictions first. |
Check the work, not just the wording
Verify facts and sources
Open the cited source, check dates and recalculate important figures. A convincing answer is not evidence.
Look for unequal effects
Ask whether a recommendation works differently for people because of disability, caring responsibilities, age, race, sex or another protected attribute.
Keep human authority real
The reviewer needs enough time, information and authority to disagree, correct or stop the outcome.
If AI affects you
1. Ask: What system was used, for what purpose, what information about you was used or inferred, and which person can review the result?
2. Record: Keep the date, notice, decision and practical effect. Keep only material you are lawfully entitled to retain.
3. Use the workplace process: Raise the issue with the nominated manager, privacy contact, safety representative, HR contact, union or worker representative as appropriate.
4. Check the applicable route: An award, agreement, privacy process, safety law, discrimination law, surveillance law, contract or policy may provide a specific pathway.
Transparency and human review are sound governance practices, but they are not universal legal entitlements for every use of AI. See the worker, applicant and contractor rights hub for jurisdiction-aware escalation options.
Common workplace moments
Meeting transcript
Check notice and permission expectations, remove unnecessary sensitive detail, verify actions and avoid treating the transcript as an infallible record.
Performance flag
Ask what data and time period were used, whether context was missing, and who has authority to review or correct the result.
AI-written draft
Confirm facts, tone, confidentiality, copyright and the accountable approver before sending or publishing it.
Sources and currency
Jurisdiction: Australia. Currency: Sources checked 27 October 2026. Legislation and workplace instruments are binding only where they apply. Regulator and government material is guidance unless stated otherwise. Standards and recommended governance practices are not legislation.