APRA to stress test banks and super funds for AI disruption in 2027

Australia's prudential regulator APRA will run its 2027 System Stress Test with AI-driven economic disruption as a named risk, the first time it has modelled that risk. The test covers banks, superannuation funds and life insurers. According to APRA's November 2026 System Risk Outlook, published on 12 November 2026, the exercise will examine two scenarios: rapid AI adoption that upends established business models, and a sharp reassessment of AI-related asset values. APRA chair John Lonsdale said major institutions were not yet where they need to be and that interruptions should be expected. For businesses that automate routine work, the practical signal is that AI suppliers are now a supervisory concern, not only an IT choice. Owners should know which AI tools their workflows depend on and whether they could move to another provider.
In short
- APRA's 2027 System Stress Test will model AI-driven economic disruption for banks, super funds and life insurers, the first time it has done so.
- The two scenarios are rapid AI adoption that upends business models and a sharp fall in AI-related asset values.
- APRA reported that 80 of the 100 largest material service providers to regulated institutions are headquartered or principally operate overseas.
- The regulator found few institutions had tested how to switch away from a critical AI provider.
- Businesses should map which automated workflows depend on which AI tools, and write down a fallback for each.
What APRA announced
APRA published its November 2026 System Risk Outlook on 12 November 2026. Coverage followed on 13 November. In the report, the regulator said its 2027 System Stress Test will explore how AI-driven economic disruption could affect banks, superannuation funds, life insurers and the broader financial system. It is the first time APRA has built AI disruption into a system-wide stress test.
The test will run two severe but plausible scenarios. In the first, rapid AI adoption disrupts established business models. In the second, confidence in AI-linked asset values falls sharply. According to the regulator's own description, the aim is to see how these shocks spread across institutions, not only inside one firm.
APRA chair John Lonsdale said Australia's financial system remains resilient, but that the risk environment is becoming less forgiving. In an interview reported by the ABC on 13 November 2026, he said major institutions were not where they needed to be in preparing for AI risk. He also said: there will be interruptions, and there will be problems.
Why the test matters for businesses that automate work
The outlook does not treat AI as a single threat. It describes four connected channels: AI used inside institutions, AI used against them through cyber attacks and fraud, shared dependencies on AI and service providers, and wider effects on employment, asset prices, competition and business models. APRA said these channels interact, so a failure in one can add stress to another.
The dependency point matters most for operations teams. Financial Standard reported on 12 November 2026 that 80 of the 100 largest material service providers reported to APRA-regulated institutions are headquartered or principally operate overseas. APRA warned that this creates possible routes for widespread disruption. A workflow that runs on one overseas AI model, one document reader or one hosted platform is exposed to the same kind of concentration the regulator is now testing.
The outlook also says few institutions had tested how to switch away from a critical AI provider. That gap is the practical one. A business may have automated invoice intake, reconciliations or customer replies without ever writing down what happens if the tool stops working for a week.
What changes in practice for a business that automates work
APRA's test applies to regulated entities. A five-person firm that is not a bank, fund or insurer is not part of the 2027 test. Still, banks, super funds and insurers buy services from many businesses. Their supplier reviews tend to move down the chain, so a smaller supplier may be asked how it depends on AI tools. The questions below are the ones an operations head can answer now.
| Question | What to write down |
|---|---|
| Which AI tools run in your workflows? | Tool name, vendor, country where the vendor is headquartered, and the task it performs |
| What happens if it stops? | The manual fallback, the person responsible, and the longest delay the business can tolerate |
| Can the work move elsewhere? | An alternative tool or manual process, tested at least once on real data |
| Where does your data sit? | Storage location, export options, and who holds copies of your records |
| Who approves changes? | A named person who checks outputs before money moves or customers are contacted |
Automation of this kind should keep a person in the loop for decisions with money, legal or customer impact. Automated steps are fastest when they handle the routine cases. Exceptions, disputed amounts and unusual requests still need a human decision. A business that removes that human check to save time has moved its risk, not removed it.
What the test does not tell you yet
The regulator has not yet published the numerical assumptions behind the two scenarios. Public coverage so far describes the scenarios in general terms only. No institution, participant list or measurement method has been set out in detail. Any business planning around the test should treat the scenarios as direction, not as a specification it can build against.
Automation also has limits that the test will not fix. An AI tool that is well documented can still produce wrong output. A workflow that is moved to a second provider needs time and testing. Results from any automated process depend on the quality of the inputs and on the owner's own figures, which only the business's own records can show.
What to watch next
The next milestone is the 2027 System Stress Test itself. The outlook does not give a date for the results. Watch for the published scenarios and for any guidance APRA issues to regulated entities on AI suppliers. APRA and ASIC have also held industry roundtables on frontier AI, and their outcomes may shape the expectations that reach smaller suppliers.
For operations heads, the practical step is to map which automated workflows depend on which AI tools this quarter. That map is the starting point for any supplier review, whether it comes from a bank, a fund, an insurer or an internal risk committee.
Sources
How AiStaffo would automate this
AiStaffo would start by listing every automated workflow the business runs, such as invoice intake, bank reconciliations, payment follow-ups and report drafts, and recording which AI tool or hosted service each step depends on. The system would connect to the accounting and document sources the business already uses and run the routine steps on a schedule, logging each action with a timestamp and the tool that performed it. Exceptions, disputed amounts and anything that touches money or customers would be routed to a named person for approval. The owner still decides which provider to use, approves any change of tool, and tests the manual fallback at least once a year. Book a free automation audit.
Questions people ask
What is the APRA 2027 stress test for AI disruption?
Does the APRA stress test apply to my business?
What did APRA chair John Lonsdale say about AI risk?
Why do AI suppliers matter for stress testing?
What should an operations head do now?
When will the APRA AI stress test results be published?
Book a free automation audit
Thirty minutes. We look at one process you run every week and tell you exactly what an AI worker would take off your desk, and what it would not.

































