Fair Work Act 2009: general protections and algorithmic adverse action
Part 3-1 of the Fair Work Act prohibits an employer or prospective employer from taking adverse action against a person because of a protected attribute — including race, colour, sex, sexual orientation, age, physical or mental disability, marital status, family or carer's responsibilities, pregnancy, religion, political opinion, national extraction, social origin, or industrial activity.
Refusing to shortlist a candidate is adverse action. When an AI screening tool filters CVs or resumes and a recruiter relies on that output without review, the agency inherits the tool's reasoning. If the model has learned to down-weight CVs with employment gaps (correlated with carer's responsibilities or disability), unusual name formats (correlated with national extraction), or older graduation dates (correlated with age), the agency faces an adverse action claim — and under section 361, the reverse onus applies: the respondent must prove the protected attribute was not a reason.
The practical takeaway: never let an AI tool auto-reject. Use it to summarise, rank, and flag — the human recruiter chooses who proceeds.
The four federal anti-discrimination Acts
Four federal statutes sit alongside Fair Work:
- Sex Discrimination Act 1984 — sex, gender identity, sexual orientation, intersex status, marital or relationship status, pregnancy, breastfeeding, family responsibilities.
- Racial Discrimination Act 1975 — race, colour, descent, national or ethnic origin.
- Disability Discrimination Act 1992 — past, present, future or imputed disability, including mental health conditions.
- Age Discrimination Act 2004 — age, including inference from graduation year or career length.
Each Act prohibits both direct discrimination (treating a person less favourably because of a protected attribute) and indirect discrimination (applying a requirement that disadvantages people with a protected attribute and is not reasonable). Indirect discrimination is where AI tools most often fall over. A model trained on an historically male-dominated industry may associate seniority cues with masculine-coded language; a tool scoring "cultural fit" from video interviews may penalise candidates with accents or disabilities. The Australian Human Rights Commission's 2026 guidance makes clear that indirect discrimination through algorithmic design is actionable.
Privacy Act 1988 and APP 6, 8 and 11
Candidate CVs, resumes, interview transcripts and reference notes are personal information, and sensitive information (health, race, religion, trade union membership) attracts higher protection. Three APPs do the heavy lifting when AI tools are involved:
- APP 6 — Use and disclosure: personal information collected for recruitment can only be used for that primary purpose, or a related secondary purpose the candidate would reasonably expect. Feeding CVs to an AI tool to generate training data for a product is not a reasonable expectation.
- APP 8 — Cross-border disclosure: before sending candidate data to an overseas AI provider, the agency must take reasonable steps to ensure the recipient does not breach the APPs — or obtain informed consent. Most commercial AI tools process data in the United States. Document the transfer in your APP 5 collection notice.
- APP 11 — Security: reasonable steps to protect personal information from misuse, interference and unauthorised access. Free consumer AI accounts, where inputs may be used for model training, generally fail this test. Paid, enterprise-tier plans (including Claude Cowork) that contractually exclude training on customer inputs are the baseline.
The Privacy Act reforms enacted through 2025 and 2026 add a statutory tort for serious invasions of privacy and strengthen automated decision-making transparency — agencies will need to tell candidates, in plain English, when AI is used in decisions that significantly affect them.
State and territory anti-discrimination overlays
Federal law is a floor, not a ceiling. Each state and territory adds attributes and sometimes stricter tests:
- NSW — Anti-Discrimination Act 1977: adds carer's responsibilities and transgender status; the NSW tribunal has been active on indirect discrimination.
- Victoria — Equal Opportunity Act 2010: includes a positive duty to eliminate discrimination so far as possible — a proactive obligation that reaches AI tool selection and bias testing.
- Queensland — Anti-Discrimination Act 1991: covers irrelevant medical record, association, and family responsibilities; updated in 2024–26 to strengthen indirect discrimination remedies.
- Western Australia — Equal Opportunity Act 1984: includes gender history, family responsibility, and religious conviction.
South Australia, Tasmania, the ACT and the Northern Territory each have their own Acts with comparable coverage. Agencies placing candidates across states must comply with the strictest applicable overlay.
Labour-hire licensing schemes
Labour-hire licensing schemes operate in Queensland, Victoria, South Australia and the ACT. Licence holders must be fit and proper, comply with all relevant workplace laws, and — in several schemes — report regularly on their compliance. Deploying AI screening or scoring tools engages the obligation in two ways. First, licence holders remain liable for discrimination and privacy breaches that flow from AI decisions. Second, workers and host employers increasingly expect disclosure of algorithmic screening; the Victorian Labour Hire Authority and the Queensland Labour Hire Licensing Compliance Unit have flagged AI transparency as an emerging compliance focus. A simple disclosure in candidate onboarding packs — "we use AI to help summarise CVs and resumes; a human recruiter makes all shortlisting decisions" — is now expected practice.
AHRI and RCSA ethical guidance
Two professional bodies set the industry benchmark. The RCSA Code for Professional Conduct binds member agencies to act with integrity, avoid discrimination, and handle candidate information responsibly — all of which apply directly to AI tool use. The Australian HR Institute (AHRI) has published guidance calling for transparency, human oversight, bias testing, and candidate recourse whenever AI is used in selection. Non-member agencies should treat both as the de facto standard a tribunal will measure them against.
Five-step compliance checklist for Australian agencies
- Human-in-the-loop by design. AI summarises and ranks; a named recruiter decides. Record the human decision in your ATS (Bullhorn, JobAdder, FastTrack360) against every candidate.
- Paid, enterprise-tier AI only. Claude Cowork, or equivalent, with contractual confirmation that candidate inputs are not used to train models. Free consumer tiers are not APP 11 compliant for sensitive recruitment data.
- Update your APP 5 collection notice and privacy policy. Disclose AI-assisted screening, the categories of data processed, the overseas jurisdictions involved, and the candidate's right to request human-only review.
- Bias test on real data, quarterly. Sample 50 CVs or resumes across demographic proxies and compare AI rankings to human rankings. Document the review — this evidence defends you against indirect discrimination claims.
- Train the team. Every consultant using AI should know the Fair Work general protections, the four federal Acts, APP 6, 8 and 11, and the agency's documented AI policy. Training records matter.
This article is general guidance for Australian recruitment professionals and is not legal advice. Agencies should obtain tailored advice from an Australian employment lawyer before deploying AI tools in high-stakes selection decisions.