Australian recruitment agencies can still use AI to screen candidates after 10 December 2026, provided the agency discloses that use in its privacy policy. The RCSA Industry Census 2025 found 39% of Australian and New Zealand recruitment professionals already use AI for process automation. The obligation changes the paperwork, not the permission.
The idea
Australia's automated decision-making transparency obligation takes effect on 10 December 2026 and it reaches AI-assisted recruitment screening, not just fully automated rejection. The Office of the Australian Information Commissioner confirmed in its Automated Decision-Making Transparency Obligation issues paper (2026) that the obligation covers computer programs which do something "substantially and directly related to making" a decision, including systems that recommend a decision or guide a human decision-maker. Johnson Winter Slattery (2026) lists AI-assisted recruitment screening among the examples in scope, which is the same layer of tooling covered in how AI-assisted sourcing shortens time to shortlist.
The obligation itself is narrow, and that is the part most agency owners get wrong in both directions. White & Case (2026) describes the requirement as a privacy policy disclosure obligation rather than a broader operational one, so an Australian recruitment agency is not being asked to justify its screening logic, prove the absence of bias, or give every applicant a right of explanation. The agency is being asked to say, in its privacy policy, what kinds of personal information its automated programs use and what kinds of decisions those programs make or assist with. The difficulty is not the writing, it is knowing the answer.
Why it matters
For an Australian recruitment agency, the commercial exposure sits in client contracts and consultant capacity rather than in regulator penalties. Staffing Industry Analysts found in its ANZ Staffing Executive Outlook 2026 that data privacy is the number one compliance concern for 70% of ANZ staffing executives, and that executive concern about AI in recruitment doubled from 20% to 45% year on year. Agencies tendering for preferred supplier panels through late 2026 will be asked how their screening works, and an agency that cannot describe it in a sentence will end up being asked to switch it off. Switching it off carries its own price, given the revenue gap between agencies that have adopted AI and those that have not.
Worth flagging which parts of an agency workflow the obligation is likely to reach:
| Where AI sits in an agency workflow | Why it is likely captured |
|---|---|
| CV parsing that ranks or scores applicants | Ranking limits which candidates a consultant ever sees, and the OAIC issues paper (2026) treats limiting employment opportunities as significantly affecting a person's rights or interests |
| AI-generated shortlists presented to a consultant | The program is substantially and directly related to the shortlisting decision even where a consultant signs it off |
| Automated knock-out questions or screening chat | The filter removes applicants on its own, with no consultant reviewing the applicant at all |
| AI matching that surfaces candidates against a live role | The match decides who gets contacted, which limits who can be considered for the role |
| Interview note-taking or call summarisation with no filtering effect | The program records rather than decides, so the decision stays with the consultant |
The commercial question underneath all of this is what the screening step is actually worth, because that is what an agency puts at risk by leaving the work until December. Take an agency with 8 consultants, each running 6 live roles a month, at ~80 applications a role, and assume 2 minutes of manual first-pass review per application. That is 160 minutes, or ~2.7 hours, of first-pass screening per role, across 576 roles a year, which comes to ~1,536 consultant hours a year. At a fully loaded consultant cost of $55 an hour, the screening step is worth ~$84,000 a year to that agency. The inputs are illustrative, so run your own role count, application volumes and hourly cost through the same arithmetic and the shape of the answer will hold. Those recovered hours feed placement speed as a direct revenue driver, which is where the screening step earns its keep.
Set against ~$84,000 of annual consultant time, the disclosure work is a mapping exercise measured in days: list every point where a program touches a candidate decision, describe the kinds of personal information those programs use, describe the kinds of decisions they make or assist with, and record where a consultant genuinely reviews the output rather than simply accepting it.
The Alvo take
The agencies that will handle 10 December 2026 cleanly are the ones that already know where their AI sits, and on the current evidence most Australian businesses do not. The National AI Centre found (2026) that only around half of Australian businesses using AI check outputs before those outputs affect a customer, with materially lower take-up of transparency practices again. Deloitte's global survey of 3,235 business and technology leaders across 24 countries found only 21% have mature governance for agentic AI (2026), and whilst that is global data rather than Australian, the read-across here is the same one the National AI Centre numbers point to, which is that adoption has run well ahead of documentation.
The pragmatic pathway is a short mapping exercise rather than a compliance project, and it is the same exercise Alvo runs as the first step of an AI readiness audit. Write down every tool in the stack that touches a candidate, mark the ones that rank, filter, score or shortlist, and confirm for each one whether a consultant genuinely reviews the output or simply accepts what the tool produces. Most agencies find a couple of tools they had forgotten were making decisions at all, usually sitting inside the applicant tracking system or a job board integration rather than in whatever the team thinks of as "the AI".
OAIC guidance is expected in September 2026 (Johnson Winter Slattery, 2026), which leaves roughly ten weeks between the guidance landing and the obligation taking effect. Agencies that map their workflow now will spend September confirming wording rather than spending December deciding what to turn off. Alvo does not provide legal advice and the policy wording itself is a job for your lawyer, but working out where AI actually sits in your workflow is an operational question, and it is the part that takes the longest.
Australian recruitment agencies can keep using AI to screen candidates after 10 December 2026. The obligation is a disclosure requirement, not a ban, and the work is a mapping exercise measured in days. Agencies that map where AI sits in their workflow now will spend September confirming wording rather than spending December deciding what to turn off.
Sources: OAIC Automated Decision-Making Transparency Obligation Issues Paper, May 2026. Johnson Winter Slattery, June 2026. White & Case, June 2026. RCSA Industry Census 2025 and Staffing Industry Analysts ANZ Staffing Executive Outlook 2026, accessed via The Access Group ANZ Recruitment Trends 2026. National AI Centre AI adoption insights, May 2026. Deloitte, April 2026.
Common questions
Can recruitment agencies in Australia still use AI to screen candidates after December 2026?
Yes. The automated decision-making transparency obligation that starts on 10 December 2026 is a disclosure requirement, not a ban. An Australian recruitment agency can keep using AI to screen, rank or shortlist candidates as long as its privacy policy sets out the kinds of personal information those programs use and the kinds of decisions they make or assist with.
Does the December 2026 privacy change apply if a consultant reviews every AI shortlist?
Yes, in most cases. The Office of the Australian Information Commissioner has indicated the obligation reaches computer programs that are substantially and directly related to making a decision, which includes systems that recommend a shortlist or guide a consultant. Having a human sign off the final decision does not by itself take the tool outside the obligation.
What exactly does a recruitment agency have to put in its privacy policy?
The privacy policy needs to describe the kinds of personal information used by the automated programs, the kinds of decisions those programs make on their own, and the kinds of decisions where the program does something substantially and directly related to a decision a person makes. It is a description of the workflow, not a technical specification.
Which recruitment tools are most likely to be caught?
Anything that ranks, scores, filters or shortlists applicants is the place to start, which usually means CV parsing and ranking inside the applicant tracking system, automated knock-out questions, screening chat, and AI candidate matching against live roles. Tools that only record or summarise, such as interview note-takers, generally do not decide anything on their own.
When does the OAIC guidance come out and how long is there to prepare?
Guidance from the Office of the Australian Information Commissioner is expected around September 2026, with the obligation commencing on 10 December 2026. That leaves roughly ten weeks between the guidance landing and the deadline. Agencies that have already mapped where AI sits in their workflow will only need to confirm wording in that window.
How much does it cost a recruitment agency to prepare?
The main cost is time rather than software. Mapping every tool that touches a candidate decision, confirming where a consultant genuinely reviews output, and drafting the policy wording is typically a couple of days of work for a small to mid-sized agency, plus a legal review of the final wording. Most agencies already own the tools involved.