APRA and ASIC Asked 380 Entities to Prove Their AI Resilience. Brokers Aren’t on That List — But Every Platform You Log Into Is

On 27 August, APRA and ASIC jointly published an information paper called Insights from the APRA-ASIC Industry Roundtables. It distilled nine roundtables held across June and July, attended by more than 600 people from more than 380 entities — banks, insurers, superannuation trustees, market infrastructure providers and the service providers sitting behind them. The regulators’ message was blunt: the financial system has done the awareness phase on frontier AI, and it is now time to produce evidence.

Five days later, at its Ignite conference in Hobart, Mortgage Choice set out what the other side of that equation looks like in the broker channel. As reported by Broker Daily and The Adviser on 2 September, chief executive Anthony Waldron told the room the group had hired its first Chief AI Officer, Thyago Liberalli, built 13 AI-native platforms and tools, and was targeting a reduction of roughly 60 to 80 per cent in the time spent on manual document processing — file renaming, tax file number redaction, the administrative sludge that sits between a signed application and a submitted file. In the same session, Waldron said the group had hired an additional Chief Information Security Officer dedicated to broker operations and was restricting access to certain external systems. “The cyber threat is not getting smaller; it is getting larger,” he said, as reported by The Adviser.

Those two events are usually filed in different folders — one regulatory, one technology. They belong in the same one. The paper that landed on 27 August is not about whether AI gives good credit advice. It is about dependencies. And the broker channel is now, unmistakably, a dependency.

What the regulators actually said — and what they did not

It is worth being precise here, because the summaries have been loose.

The APRA-ASIC paper is concerned with frontier AI as a threat and a concentration risk. It is not a ruling on AI-assisted advice, it does not create a licensing condition, and it says nothing about mortgage broking specifically. Read it as a resilience document, not an advice document.

Its findings clustered around five themes. Cyber fundamentals — asset visibility, patching, identity controls, monitoring, backup integrity — matter more, not less, when attackers can move faster. Governance is a live vulnerability in its own right: the paper records participants identifying weak escalation pathways as potentially as disruptive as a technical control failure. Defensive AI is promising but immature, and the regulators noted that “governed, measurable and scalable defensive AI capability remains limited.” Shared dependencies on cloud providers, SaaS platforms, managed service providers and AI model providers create the possibility that one incident cascades across the sector. And collective information-sharing improves everyone’s odds.

The operative word across the paper is evidence. The regulators say they expect entities to demonstrate that decisions, escalation pathways and recovery arrangements actually function at the speed the threat now demands — tested before disruption, not asserted after it. They signalled this will remain a heightened supervisory focus. They also set out expectations on third-party dependencies: know which providers support critical operations, understand where a single failure hits multiple entities, and keep tested fallback arrangements with clear accountability between the entity and the provider.

That last paragraph is the one that reaches you.

Why a paper addressed to prudentially regulated entities lands on your desk anyway

Brokers are not APRA-regulated. Most brokers operate as credit representatives rather than licensees. Nothing in this paper creates a direct obligation for a broker writing files in Wagga or Wangaratta.

But you are a third party in somebody else’s dependency map, and so is every platform you use. Your aggregator’s CRM, your lender’s origination portal, your servicing calculator, your policy search tool, your document collection app — these sit inside the assurance perimeter of entities that were in those roundtables, or supply entities that were.

When a regulated entity is told to map its dependencies, test fallbacks and maintain clear accountability with providers, that requirement does not stop at the entity’s own boundary. It travels down the chain. In practice, brokers should expect it to arrive as: vendor assurance questionnaires passed on by aggregators, tightened authentication and session controls, restricted or revoked integrations between third-party tools and lender systems, more prescriptive AI-use policies from aggregators and licensees, and slower approval cycles for bringing a new tool into the workflow.

Mortgage Choice restricting access to certain external systems, announced in the same week, is what that looks like at the coalface. It is not an outlier. It is the leading edge.

The obligation that is already yours has not moved an inch

Here is the part that gets muddled when a group announces a 60 to 80 per cent reduction in processing time.

The best interests duty sits in sections 158LA and 158LE of the National Consumer Credit Protection Act 2009, and has applied to mortgage brokers since 1 January 2021. ASIC’s guidance in Regulatory Guide 273 is explicit about where compliance is demonstrated from. At RG 273.21, ASIC states: “We expect that evidence of compliance with the best interests obligations will come predominantly from the broker’s records.” At RG 273.162, it says: “As a mortgage broker, we expect you to keep records of how you have acted when providing credit assistance. This includes records of inquiries you make into the consumer’s circumstances, and the consideration, investigation and assessment of the products you recommend.”

RG 273.29 makes the framing clear: “We consider that the processes and actions of the mortgage broker are a key consideration when determining whether the best interests duty has been complied with.”

Read those three together and the implication for AI-assisted workflow is straightforward. If a policy search tool returns a ranked list of lenders that fit a scenario, that output is an input to your consideration. It is not a record of your consideration. The file still has to show what you inquired into, what you assessed, and why you recommended what you recommended.

The same logic runs the other way. If an AI agent drafts a file note, the broker owns what it says. A drafted note that asserts an inquiry you did not actually make is a worse compliance position than no note at all.

This is general information, not compliance advice, and the practical answer will differ by licensee. Brokers should confirm with their aggregator’s compliance team or licensee how AI-generated content is expected to be recorded, reviewed and retained in their particular file structure.

The concentration question the channel is not asking about itself

The paper’s fourth theme — shared dependencies creating systemic risk — deserves a broker translation.

The direction of travel in the channel is consolidation onto integrated platforms. Policy search, document processing, client engagement, marketing, compliance review and CRM increasingly sit inside one environment, and that environment increasingly sits on one cloud provider and one large language model provider. Mortgage Choice’s build is candid about this: its tooling includes agents built on Google Gemini for tasks including document verification and meeting assistance.

That is a reasonable commercial choice and it delivers real time savings. It is also, precisely, the concentration the regulators described. The broker-level version of the question is not abstract: if the model provider behind your policy search and your document processing has a bad afternoon, what can you still do? Can you write a file? Can you find a lender’s construction policy? Can you get a client’s documents collected and named?

Most brokers have never answered that, because they have never had to. Building a two-line fallback for each critical tool is an afternoon’s work and it is the cheapest resilience available to a small business.

What to review this week

A practical starting point is a one-page AI and dependency register for the business. It does not need to be sophisticated; it needs to exist.

For each tool in the workflow, record five things: what it is, what it touches (client personal information, credit data, the recommendation itself, or just marketing), who approved it and under what policy, what the client file shows when the tool has been used, and what you do if it is unavailable for a day.

Then work through six checks:

1. Identify every tool currently touching client data that was adopted by an individual broker rather than approved by the licensee. Shadow adoption of consumer AI tools is the most common gap and the easiest to fix.
2. Read your aggregator’s current AI use policy. If you cannot find one, ask. Several aggregators have updated theirs in 2026 and the update may not have been well socialised.
3. Pull three recent files where a policy or servicing tool narrowed the lender shortlist. Check whether the file evidences your reasoning or only the tool’s output.
4. Confirm how tax file numbers, identity documents and bank statements are handled by any automated document tool — where they are processed, where they are stored and for how long. Redaction being automated does not change who is accountable for the data.
5. Write the fallback line for each critical tool. “If the CRM is down I work from the exported pipeline CSV and the shared drive” is a plan. “It won’t go down” is not.
6. Nominate who in the business decides to stop using a tool mid-incident. The APRA-ASIC paper’s point about escalation authority scales down to a three-person brokerage without much modification: if nobody has the call, the call does not get made.

What to watch next

The regulators have signalled continued supervisory attention, so the practical flow-through for brokers is likely to be indirect and gradual — vendor assurance and platform controls rather than anything addressed to the broker channel by name. Watch for aggregator AI policies being reissued, for lenders tightening what third-party tools may connect to their portals, and for professional indemnity insurers starting to ask how AI is used in the advice process.

The strategic read is simpler than the compliance read. The channel is adopting AI faster than it is governing it, and the regulators have just told the rest of the financial system that awareness without evidence no longer counts. Brokers who can show, on a file, that the tools accelerated the work without ever making the recommendation will be in a materially stronger position than brokers who cannot — regardless of whether anyone from ASIC ever asks.

The time savings are real. The accountability did not move.