A candidate misses out on a role because an automated score places them below a shortlist threshold. If the organisation cannot explain what influenced that score, show that it relates to the job, or demonstrate meaningful human review, the hiring process may be difficult to defend. For Australian employers, the question is not simply whether AI scoring legal. It is whether the way AI is used is fair, lawful, valid and proportionate to the hiring decision being made.
AI can improve recruitment when it helps teams process assessment data consistently, identify relevant patterns and focus human attention where it is needed. It can also create legal and reputational risk when its outputs are treated as objective truth rather than one input in a well-governed selection process.
When is AI scoring legal in Australia?
There is no single Australian law that makes AI scoring in recruitment automatically legal or illegal. Legality depends on the tool, the data it uses, the decision it influences and the safeguards around it.
An AI-enabled hiring process may need to meet obligations under privacy law, Commonwealth and state or territory anti-discrimination law, workplace law and, in some circumstances, surveillance and biometric information requirements. Government agencies and organisations operating in regulated sectors may also face additional procurement, record-keeping and public sector obligations.
The practical test is straightforward: can the employer show that the scoring process assesses capabilities genuinely relevant to the role, operates fairly across candidate groups, handles personal information appropriately and does not replace accountable human judgement? If the answer is unclear, implementation should pause until the process can be properly evaluated.
AI scoring is not one thing. It may involve ranking candidates after a cognitive ability test, identifying themes in asynchronous video interview responses, flagging inconsistencies in application data, or producing a job-fit report from several assessment measures. Each use case carries a different level of risk. Scoring that helps a recruiter organise a high-volume applicant pool is not equivalent to scoring that automatically rejects a candidate from a leadership role.
Privacy starts before the score is produced
Recruitment data is personal information, and it can be highly sensitive. Candidates may provide employment history, assessment results, video responses, identification details and information that can reveal disability, cultural background or other protected attributes. Where biometric information is collected or generated, the risk profile may increase further.
For organisations covered by the Privacy Act 1988, candidate information should be collected only where it is reasonably necessary for recruitment, handled in line with the Australian Privacy Principles and protected with appropriate security controls. The employee records exemption should not be treated as a shortcut in recruitment: applicants are not yet employees.
Clear candidate information is essential. People should understand what information is being collected, why it is being assessed, whether AI contributes to scoring or recommendations, who will access the results, how long information will be retained and whether data is stored or processed overseas. A vague statement that technology may be used is unlikely to build confidence or provide meaningful transparency.
Consent can be relevant, particularly for particular assessment or recording activities, but it is not a complete governance strategy. Candidates may feel they have little practical choice when applying for work. Employers should therefore be able to justify the collection and use of information independently of a tick-box consent mechanism.
Discrimination risk is often hidden in the inputs
AI systems do not need to ask about age, sex, race, disability or other protected characteristics to produce discriminatory outcomes. Bias can enter through historical hiring data, proxy variables, language patterns, training labels or design choices that favour people who have previously succeeded in a particular workplace.
For example, a model trained on past high performers may learn preferences embedded in earlier recruitment decisions rather than the real requirements of the role. A video analysis tool may also produce uneven results for candidates with disability, different communication styles, accents or limited access to high-quality technology. A score may look precise while concealing weak or uneven measurement.
Australian anti-discrimination law focuses on outcomes as well as intent. Employers should consider whether a scoring method could disadvantage people with protected attributes, whether that disadvantage is connected to a genuine job requirement, and whether reasonable adjustments are available. This is particularly relevant where timed assessments, video interviews, gamified tasks or online proctoring are involved.
The solution is not to abandon structured assessment. Properly designed assessments can reduce the inconsistency of intuition-led hiring. The issue is whether the assessment measures relevant capabilities in a consistent way, and whether the AI layer is adding sound interpretation rather than amplifying noise.
Validity matters more than novelty
A useful hiring score needs an evidence base. Before using AI to rank or screen candidates, employers should define the capabilities that predict performance in the specific role. Those capabilities might include numerical reasoning, attention to detail, safety judgement, customer communication, technical knowledge or leadership potential. They should not be confused with a system’s ability to detect patterns in large volumes of data.
This is where validated psychometric and skills assessments have a practical advantage. They provide a defined construct, standardised administration and evidence about what the result represents. AI can then help present results, highlight patterns or support reporting, while the underlying measure remains tied to job-relevant criteria.
By contrast, a black-box score based on unclear behavioural signals is difficult to defend. If a vendor cannot explain what is being measured, how the score has been tested, which populations were included in validation, and what limitations apply, the employer is taking on unnecessary risk.
It also depends on the hiring context. A lower-risk score that supports recruiter prioritisation in a large graduate campaign may be acceptable with careful monitoring. A score that determines whether an applicant progresses for a senior, clinical, safety-critical or public-facing role warrants greater scrutiny, stronger documentation and more senior human review.
Human oversight must be real
Human oversight is not satisfied by placing a recruiter at the end of an automated workflow and asking them to click approve. The reviewer needs enough information, authority and time to question the output.
A defensible process gives decision-makers access to the relevant assessment evidence, not merely a red-amber-green ranking. It allows them to account for context, reasonable adjustments and information that the system could not fairly assess. It also creates a way to challenge an outcome where a score appears inconsistent with the candidate’s demonstrated capability.
For high-impact decisions, employers should avoid fully automated rejection. A candidate should not be excluded solely because an algorithm generated a low score, particularly where the scoring method relies on opaque criteria or sensitive data. Human judgement should be structured, however, rather than used to reintroduce personal bias. Decision-makers need clear selection criteria and documented reasons for departures from standard recommendations.
A practical governance standard for AI scoring
Before deploying an AI-enabled scoring tool, HR and talent teams should complete a focused review with their privacy, legal, technology and people leaders. The aim is not to create bureaucracy. It is to ensure the tool improves the decision rather than creating an unexamined risk.
A sound review should cover four areas:
- Job relevance: Define what the score measures and establish how it relates to success in the role.
- Fairness: Test for potentially adverse outcomes across relevant candidate groups and provide reasonable adjustments where required.
- Transparency and privacy: Give candidates clear information, minimise data collection, set retention periods and assess data hosting and vendor access arrangements.
- Accountability: Set human review points, document decision rules, monitor outcomes and nominate an owner responsible for responding to issues.
Vendor due diligence is central to this process. Employers should ask how the model was developed, what data it uses, whether customer data trains the system, how outputs are monitored, how errors are corrected and whether independent validation is available. They should also understand whether the provider is making recommendations, rankings or decisions on the employer’s behalf.
Documentation matters because recruitment decisions may need to be explained months later. Keep records of the role analysis, assessment rationale, candidate communications, reasonable adjustments, reviewer guidance and monitoring results. This creates a more defensible process and gives the organisation useful evidence for improving future hiring campaigns.
Use AI to strengthen judgement, not avoid it
The strongest use of AI in recruitment is rarely a system that claims to identify the best person with no human involvement. It is technology that helps employers apply relevant, validated evidence more consistently across a large candidate pool.
RightPeople’s approach reflects this principle: scientifically grounded assessment data, practical reporting and expert interpretation can help hiring teams make faster decisions without losing sight of fairness and job relevance. AI-enhanced reporting can be valuable when it clarifies evidence for a recruiter or hiring manager. It should not become a substitute for the employer’s responsibility to make a considered decision.
Candidates are increasingly alert to how their information is used, and they are right to expect a fair process. Employers that can explain their assessment methods, provide appropriate support and apply meaningful human review will be better placed to earn trust as well as reduce risk. The goal is not to make hiring less human. It is to make human decisions more consistent, evidence-based and defensible.
