AI in Australian financial services

Start here, because it is the thing most compliance summaries get wrong by implication. As things stand there is no AI-specific legislation in force in Australia. What governs AI in financial services is the existing stack: the Privacy Act, the licensee obligations in the Corporations Act, the Australian Consumer Law and the ASIC Act, and APRA’s prudential standards for regulated entities.

In late 2024 the government consulted on introducing mandatory guardrails for AI in high-risk settings. A very large amount of published compliance content still describes those guardrails as forthcoming, and treats their arrival as a planning assumption.

That is no longer accurate. The Department of Industry, Science and Resources’ own consultation page now states that the government will not proceed at this time with the previous proposals to introduce mandatory guardrails, with the feedback folded into a national AI plan instead. Separately, Treasury’s final report on its review of AI and the Australian Consumer Law, dated 3 October 2025, concluded that the Consumer Law can generally handle AI and recommended no new binding AI legislation.

The Privacy and Other Legislation Amendment Act 2024, Act No. 128 of 2024, assented on 10 December 2024, introduced the automated-decision obligation. It is worth being precise about its form, because almost everyone describes it as a new section of the Privacy Act and it is not. It inserts new Australian Privacy Principles 1.7, 1.8 and 1.9 into clause 1 of Schedule 1, with a consequential amendment to the enforcement provision in section 13K. Anyone citing a section number for this obligation is citing something that does not exist.

APP 1.7 is triggered where an entity has arranged for a computer program to make, or to do a thing that is substantially and directly related to making, a decision; where that decision could reasonably be expected to significantly affect the rights or interests of an individual; and where personal information is used in the program’s operation. Note the reach of “ substantially and directly related”: a program that scores or triages an application without issuing the final decision can still be caught.

Where it applies, APP 1.8 requires the privacy policy to set out the kinds of personal information used, the kinds of decisions made solely by the program, and the kinds of decisions where a substantially-and-directly-related thing is done by the program. APP 1.9 gives examples, including decisions affecting an individual’s rights under a contract and their access to a significant service or support, which is squarely lending and insurance territory.

The obligation commences on 10 December 2026. It is not law today, and a surprising amount of published material marks it as already in force, in at least one case while the same page’s own body text gives the correct 2026 date.

The check anyone can run: the current compilation of the Privacy Act stops at APP 1.6. There is no APP 1.7 in the in-force text, and the in-force Act contains no occurrence of the phrase “automated decision” or “computer program” anywhere in it.

One forward-looking detail worth planning around: when it does commence, it applies to decisions made after that date even where the system and the underlying personal information pre-date it. An existing model is not grandfathered. Enforcement runs through the infringement and compliance notice machinery in section 13K.

APP 1.7 to 1.9 is a privacy-policy disclosure obligation. It creates no right to human review, no right to an explanation of an individual decision, and no right to object to automated processing. Reading it as Australia’s Article 22 overstates it considerably.

The broader GDPR-equivalence assumption fails elsewhere too: Australian privacy law has no legitimate-interests basis and no general right to erasure. Mapping a European compliance programme onto Australia one-to-one will produce both false obligations and missed ones.

As for what constrains AI under the Privacy Act today, the OAIC leans on APP 3, governing collection, and APP 6, governing use and disclosure, rather than on the security and access principles that summaries tend to reach for first.

ASIC’s principal AI publication is REP 798, “Beware the gap: Governance arrangements in the face of AI innovation,” dated 29 October 2024. It reviewed 23 AFS and credit licensees covering 624 AI use cases as at December 2023, with 61 per cent planning to increase their use of AI.

Its stated position is that the regulatory framework for financial services and credit is technology neutral, and that several existing obligations already bear on the safe and responsible use of AI, naming in particular the general licensee obligations, the consumer protection provisions, and directors’ duties. In practice that means the obligation to provide services efficiently, honestly and fairly, the design and distribution obligations, unconscionable conduct, adequate resources, and risk management. There is no AI-specific regulatory guide.

It would be wrong to read that as ASIC considering the matter settled. Its chair has said publicly that just because existing regulation can apply to AI, that does not mean there is nothing more to do.

For APRA-regulated entities, CPS 230 on operational risk management commenced on 1 July 2025, with pre-existing service-provider contracts transitioning at the earlier of their renewal or 1 July 2026. It applies to ADIs, general, life and private health insurers, and RSE licensees. CPS 234 on information security has applied since 1 July 2019 and carries a 72-hour material incident notification requirement.

Neither standard mentions AI, and neither does the related guidance. AI is caught indirectly, through the critical operations and tolerance-level machinery and through the material service provider obligations, which take in core technology services. APRA’s letter to industry on AI, dated 30 April 2026, states that its principle-based prudential framework is technology and vendor agnostic and that it is not proposing to introduce additional requirements at this stage.

The small business exemption in section 6D survives untouched. An entity with annual turnover of $3 million or less is generally outside the Act, and the 2024 amendments did not alter that threshold. A smaller lender or broker may therefore have no Privacy Act obligations at all, including no APP 1.7 obligation when it commences.

One narrowing to note: from 1 July 2026, entities brought into the anti-money-laundering regime are covered in respect of their AML and CTF-related handling of personal information, which is a targeted carve-in rather than a removal of the exemption.

On breaches, the notifiable data breach scheme requires reasonable steps to complete an assessment within 30 days, and a statement to the Commissioner as soon as practicable after the entity becomes aware.

The Australian obligation that is actually coming is a disclosure obligation, and disclosure is only as accurate as your inventory. APP 1.8 asks what kinds of personal information a program uses and which decisions it makes solely or near-solely, which is a question about system provenance rather than about model internals. Nova AIS puts a permission model and an immutable audit trail under the systems built on it, so that inventory is derived from the record rather than assembled by survey before a deadline.

The “substantially and directly related” wording is the part worth designing around early. A scoring or triage step that never issues the final decision can still fall inside the rule, so the useful question before December 2026 is not just which systems decide, but which systems materially shape a decision.