Updated 17 October 2026. We have now read the updated APP 1 Guidelines and the flowchart, which this story first mentioned without quoting. They go further than the fact sheet on staff use of AI chat tools, on decisions that land differently for vulnerable people, and on writing a disclosure that can be challenged. The new section is What the full guidelines add.
In August we reported that a privacy rule starting on 10 December 2026 would require
organisations to disclose automated decisions, and that the regulator had not yet said what
counts. On 30 September the Office of the Australian Information Commissioner said. Its new fact
sheet, a supplement for government agencies, a flowchart and an update to its APP 1 Guidelines
answer the open question, and the answer is wide. The OAIC says the resources reflect the
feedback received from 90 written submissions
.
What counts as a computer program: nearly everything
The fact sheet's list of what a computer program includes runs from pre-programmed
rule-based processes
to artificial intelligence and machine learning processes
,
software, apps or word-processing tools
, and generative AI used to generate text,
images, videos, code or synthesis, including chatbots
. There is no carve-out for
old-fashioned software. A spreadsheet rule that sorts applicants is in the same category as a
large language model.
A human in the loop does not take you out of it
The point most organisations will need to hear: A decision may be within scope of the
transparency obligation even where a computer program output does not replace the entire
decision-making process or is subject to human review.
The OAIC goes further for machine
learning and generative AI. It considers their outputs, when used to make decisions that
significantly affect people, would generally fall within scope of the transparency
obligation unless subject to extensive human oversight and control
.
It also defines the two words the test turns on. Substantially means where it is a key
factor in facilitating the human's decision making
, and Directly means where the thing has
a direct connection with making the decision
.
The phrase that decides scope, now defined
Our August piece said everything turned on whether a decision could reasonably be expected
to significantly affect the rights or interests of an individual
. The fact sheet now says
significantly means the impact of the decision or thing must be more than trivial and must
have the potential to considerably influence the circumstances or outcomes for the individual
concerned
, and that vulnerability matters: a decision may have a greater impact on people
experiencing vulnerability than on the general population.
It also lists decisions it would generally consider in scope. Among them: facial recognition for watchlist matching in a shop or stadium, recruitment software that sorts candidates, personalised pricing for significant goods, programs that prioritise health or disability services, eligibility for government benefits, housing, education places, loans, credit and insurance, AI reports used to rank staff or set bonuses, and immigration and border processing.
And where an organisation is unsure, the regulator's instruction is to disclose: entities in
doubt should take a cautious approach and include information in their APP privacy policy
.
Who carries it, and what can be withheld
The obligation stays with the organisation making the decision, not the software vendor, though
the OAIC says vendors should provide clear, high-level information
about how their products
can be used so customers can comply. Commercially sensitive information and trade secrets are out of
scope, with a limit the fact sheet states plainly: information about using a program for significant decisions is not shielded just because it results in exposure to ridicule, embarrassment or public criticism
.
Government agencies get a supplement linking the rule to their existing Freedom of Information
publishing duties. It says agencies should publish the types of automated decision-making they use,
not just that AI is being used by the agency
.
What the full guidelines add
The updated APP 1 Guidelines (version 2.0, dated 30 September) set out the same tests at greater length, with worked examples. Two points go further than the fact sheet, and one sharpens it.
An employer that lets staff use a chat tool for promotions can be in scope. The
rule applies where an organisation has arranged for a computer program to make or inform a decision.
Among the examples the OAIC says would generally count as arranged for is an employer that
permits or directs employees to use an AI chat tool to draft performance assessments that
determine promotion decisions
. The other tests still apply: the decision must use personal
information and significantly affect someone, and the program must be substantially and directly
related to it. The guidelines also say organisations should
actively identify, assess and keep oversight over the third-party computer programs that they use to
make a decision, both during and after procurement
.
An output that differs for a vulnerable group is taken to be significant.
Organisations should consider whether the decision in question produces an output that is
materially different for a vulnerable cohort of individuals when compared to the general Australian
population
. Where it does, the guidelines say the decision should be taken to significantly
affect the rights or interests of an individual
. Their non-exhaustive list of significant rights
or interests includes access to telecommunications, essential utilities and physical assistance or
support, none of which the fact sheet mentions.
The disclosure should be open to challenge. The fact sheet asks for information
framed so that it and the decision can be reviewed
. The guidelines put it more strongly:
framed in a way that allows the information and the decision to be challenged or contested
.
One worked example makes the human-review point concrete. A scholarship panel that receives a
computer-generated ranking of applications, and can disregard the ranking
, is still in
scope, because the panel considers the ranking even though they may choose to ignore it
.
Our view
This is broader than a reading limited to AI would suggest, and clearer than the rule's own words. The two exits people tend to reach for, "a human signs off" and "it is just software", are both closed by the regulator's own words. With just under ten weeks to go, an organisation that has not inventoried its decision tools is behind. And the list of in-scope decisions reads like a map of where automated decisions already touch people most: hiring, pricing, credit, insurance, benefits and borders. The chatbot example in the full guidelines is the one an inventory of purchased software would miss: on the OAIC's list, an employer that permits or directs staff to draft promotion assessments with a chat tool has arranged for a computer program, whether or not it bought the tool, and if the other tests are met it owes the disclosure.
How we did this
We read the OAIC's announcement and both fact sheets in full and quote them directly. The flowchart and the updated APP 1 Guidelines were released alongside them; we first published without quoting them, then on 17 October read the flowchart and the guidelines' section on the new obligation (paragraphs 1.35 to 1.78) and added what they say beyond the fact sheet. We checked the fact sheet's text on 17 October before saying the guidelines go further. The list of in-scope decisions is the OAIC's, summarised in our words with every item kept. Fact sheets are the regulator's view of the law, not the law, and a court could read the obligation differently. "Just under ten weeks" is our count from 3 October to 10 December. This follows our 18 August story on the same obligation. We did not contact the OAIC. The “our view” paragraph is opinion based on the documents quoted.
Sources
Method note. Written substantially by AI and run by people, working from the primary sources below. 8 quotes here were verified by script against copies of the sources we archived. How AI Geek works.
- OAIC, New resources on transparency for use of AI and automated decision-making (published 30 September 2026, read 3 October 2026): the commencement date, the resources released and the 90 submissions.
- OAIC, APP 1.7-1.9 Transparency Obligation Fact Sheet (PDF, 11 pages, September 2026, read 3 October 2026): what counts as a computer program, human review, substantially and directly, significantly, vulnerability, the in-scope decisions list, the in-doubt instruction, vendors, and the commercial-in-confidence limits.
- OAIC, Supplementary Fact Sheet for Government Agencies (PDF, 5 pages, read 3 October 2026): the link to Information Publication Scheme duties and what agencies should publish.
- OAIC, APP Guidelines, Chapter 1: APP 1, Open and transparent management of personal information (version 2.0, updated 30 September 2026, read 17 October 2026, paragraphs 1.35 to 1.78): the chat-tool example, oversight of third-party programs, the vulnerable-cohort test, the list of rights and interests, contestability, and the scholarship example.
- OAIC, APP 1.7-1.9 Transparency Obligation Flowchart (PDF, 1 page, read 17 October 2026): the same tests as the fact sheet, as a decision tree; not quoted.
Spotted an error? Tell us and we will check it against the sources and log the outcome here.