Privacy and data
AI for HR in Australian small businesses: what’s safe to use
AI assistants are safe for most HR drafting and research when you keep personal information out or use a business tool you have checked. The risk sits in what you enter (Privacy Act APP 6 and APP 8), the kind of information (health, union and other sensitive information), and decisions about people (discrimination law and, from 10 December 2026, privacy policy disclosure).
Updated . General information, not legal advice.
Which HR tasks can an AI assistant help with?
Drafting letters, policies and position descriptions, summarising long documents, and getting a first answer to an award or leave question are the lowest-risk uses, provided a person checks the result before it is used.
| Task | Risk | What to watch |
|---|---|---|
| Drafting a policy, job ad or template letter | RiskLow | What to watchNo personal information needed. Check the law it states against the Act or award. |
| Summarising an award clause or a legislation page | RiskLow to medium | What to watchSummaries can be wrong. Read the clause itself before paying anyone. |
| Drafting a letter to a named employee | RiskMedium | What to watchEnter only what the letter needs. Leave out health and other sensitive details. |
| Summarising a grievance, investigation or medical certificate | RiskHigh | What to watchSensitive information and an output that becomes part of a person’s file. |
| Ranking applicants or flagging staff for performance action | RiskHigh | What to watchDiscrimination risk and, from 10 December 2026, privacy policy disclosure. |
An assistant can be confidently wrong about pay rates and entitlements. Check any figure against the award or the legislation.
Does the Privacy Act apply to my business?
A business with annual turnover of $3 million or less is generally a small business operator outside the Australian Privacy Principles, unless an exception applies. The tax file number rules apply to every employer that holds TFNs.
The $3 million test is in section 6D of the Privacy Act, and exceptions bring some small businesses back in, such as health service providers. Anyone who holds a record containing tax file number information is a file number recipient (s 11) and must follow the TFN rules (s 18), whatever their turnover. See employee records: privacy and security for the exceptions.
If you are covered, the employee records exemption (s 7B(3)) is narrower than many employers assume. It covers acts directly related to a current or former employment relationship and an employee record. The OAIC says it does not cover unsuccessful job applicants or volunteers, and is unlikely to apply to organisations providing HR services under contract to an employer. An AI tool supplier handling your staff data is in that second group.
What do APP 6 and APP 8 mean for AI tools?
APP 6 limits using or disclosing personal information for a purpose other than the one it was collected for. APP 8 requires reasonable steps before disclosing it to a recipient outside Australia, and section 16C can make you liable for that recipient’s breach.
The OAIC’s guidance on commercially available AI products says entering personal information into a publicly available chatbot is a disclosure to the tool’s owner. It must fit the original purpose of collection, or an exception such as consent or the individual’s reasonable expectation of a related secondary purpose (directly related, for sensitive information).
- APP 6: an employee’s address was collected to pay and contact them, not to feed a third-party tool. Ask whether the task fits that purpose.
- APP 8: if the tool processes or stores data overseas, you must take reasonable steps to make sure the recipient doesn’t breach the APPs. Under s 16C, a breach by that overseas recipient can be treated as your breach.
- APP 10 and APP 11: outputs about a person must be accurate, and prompts, files and outputs must be protected like any other record.
Which employee information needs extra care?
Sensitive information under section 6(1) of the Privacy Act includes health information, racial or ethnic origin, trade union membership, sexual orientation, religious beliefs, criminal record and biometric information. Collecting it generally needs consent (APP 3.3).
As a matter of best practice, the OAIC recommends not entering personal information, and particularly sensitive information, into publicly available generative AI tools. In HR the common sensitive items are medical certificates, workers’ compensation files, union membership on payroll deductions and criminal history checks.
AI outputs count too. The OAIC says information an AI tool infers or produces about an identifiable person, even if it is wrong, is personal information that must be handled under the APPs.
Can AI-assisted decisions about staff breach discrimination law?
Yes. The Fair Work Act bans adverse action against employees and prospective employees because of attributes such as age, sex, race, disability, pregnancy or carer’s responsibilities (s 351), and the employer answers for the decision whatever tool helped make it.
A tool that scores performance, summarises sick leave or screens CVs can lean on a protected attribute without anyone intending it. Keep a person accountable for every decision about hiring, discipline, pay or dismissal, and be able to explain the reasons without relying on the tool. The Fair Work Ombudsman lists not hiring someone as adverse action.
From 10 December 2026, organisations covered by the Privacy Act must also say in their privacy policy when a computer program uses personal information to make, or substantially and directly help make, decisions that could significantly affect people (APP 1.7). See automated decisions and privacy disclosure.
What records rules apply when AI is involved?
Fair Work employee records must be kept for 7 years (s 535), and the Fair Work Ombudsman says they must be legible and in English. An AI summary does not replace the record itself.
Keep the original records, not just a summary. If an assistant drafts a warning letter or an investigation summary, the final version you send or file is the record, and it must be accurate. Note which tools were used for decisions about people so you can answer a complaint or a regulator later.
If a tool exposes employee information to the wrong people, the Notifiable Data Breaches scheme may apply. A covered business that suspects an eligible data breach must take reasonable steps to finish its assessment within 30 days (s 26WH).
What should I check before putting employee data into a tool?
Check where data is stored and processed, whether inputs are used to train models, how long data is kept, who can access it, and whether sign-in needs multi-factor authentication.
| Question | Why it matters |
|---|---|
| Where is data stored and processed, including backups and any sub-processors? | APP 8 and s 16C if any of it leaves Australia. |
| Are prompts, files or outputs used to train or improve models? Can that be switched off? | The OAIC flags training use as a disclosure risk under APP 6. |
| How long is data kept, and can we delete it? | APP 11.2 requires reasonable steps to destroy or de-identify information you no longer need. |
| Who at the supplier can see our data, and is access logged? | APP 11.1: protection from unauthorised access. |
| Does sign-in support multi-factor authentication and separate admin roles? | Multi-factor authentication is one of the ASD’s Essential Eight strategies. |
| Will the supplier tell us promptly about a breach? | You need the facts to assess an eligible data breach within 30 days. |
The OAIC warns against treating due diligence as ‘set and forget’: review the tool, its settings and staff use regularly.
What is a short checklist for using AI in HR safely?
Approve the tools, keep sensitive information out of unapproved ones, check every output, keep people accountable for decisions, and write the rules down in a policy.
- List the AI tools staff already use, including built-in features.
- Approve business versions only, with training on your data switched off and storage where your privacy policy says it is.
- Ban TFNs, bank details, health and other sensitive information in any tool not approved for them.
- Remove names and identifiers when the task doesn’t need them.
- Have a person check facts, figures and tone before anything is sent or filed.
- Never let a tool make the final call on hiring, discipline, pay or dismissal.
- Turn on multi-factor authentication for every account that can reach HR data.
- Update your privacy policy before 10 December 2026 if a program shapes significant decisions.
- Write it into a policy. See how to write a workplace AI use policy and using AI assistants with HR data.
Workfloor is hosted in Sydney, its AI processing runs in Australia only, and customer content is never used to train AI models.
Questions
Can I use a free AI chat tool to draft an employee warning letter?
You can draft the structure, but the OAIC recommends not entering personal information, especially sensitive information, into publicly available AI tools. Add names and details yourself after drafting, or use an approved business tool.
Does the employee records exemption let me put staff files into any AI tool?
No. It only covers acts directly related to a current or former employment relationship and an employee record, it doesn’t cover applicants or volunteers, and the OAIC says it is unlikely to cover a contractor providing HR services to you.
My business turns over less than $3 million. Do any of these rules apply?
The APPs generally don’t, unless an exception applies, but the TFN rules, Fair Work record-keeping and discrimination laws do. The OAIC’s practices are still a sensible standard.
Sources
- OAIC: Guidance on privacy and the use of commercially available AI products
- OAIC: Employee records exemption
- Privacy Act 1988 (Cth), current compilation (ss 6(1), 6D, 7B(3), 11, 16C, 18, 26WH; APPs 3, 6, 8, 11)
- Privacy and Other Legislation Amendment Act 2024, s 2 and Schedule 1 Part 15
- OAIC: New resources on transparency for use of AI and automated decision-making (30 September 2026)
- Fair Work Act 2009 (Cth), current compilation (ss 351, 535)
- Fair Work Ombudsman: Record-keeping
- Fair Work Ombudsman: Protection from discrimination at work
- Australian Signals Directorate: Essential Eight explained
Checked against these sources on 12 November 2026. Spotted something out of date? Tell us.