AI Evidence for APRA-Regulated Entities · Triodian

Australia · APRA

APRA has asked boards to oversee AI inside their providers. The record does not exist yet.

APRA's letter of 30 April 2026 expects boards to oversee AI within appetite, including inside third-party dependencies, with clearly defined triggers. It named reliance on vendor summaries as the gap. The AI Use Assessment shows what each provider can actually evidence, before APRA asks.

What applies

The letter set no date. The standards behind it already apply.

APRA's letter says existing prudential standards apply to AI risk, without naming them. The mapping below is our reading; counsel grades what is a requirement and what is an expectation.

InstrumentWhat it asksWhat it means for evidence
APRA letter to industry on AI, 30 April 2026Boards oversee AI within appetite, including inside third-party dependencies, with clearly defined triggers. Stronger supervisory action where that is not done. No date set.An appetite with testable limits and triggers, and a record each time a trigger fires.
CPS 220 Risk ManagementThe board owns the risk appetite and its triggers.AI inside providers has to sit inside that appetite, not beside it.
CPS 230 Operational Risk ManagementReaches agreements with material service providers and the chain supporting them. Existing agreements must comply at renewal, and by 1 July 2026 at the latest.Tolerances for critical operations, including those a provider runs with AI.
CPS 234 Information SecurityControls over information assets managed by third parties, tested in proportion to the threat. Material incidents notified within 72 hours.Control testing that reaches AI in the chain, and a record that exists at the time of an incident.
Financial Accountability RegimeAccountable persons answer for reasonable steps.A record of what was asked, found and decided, before APRA or ASIC asks.
Privacy Act, APP 1.7, from 10 December 2026Privacy policies must disclose substantially automated decisions that significantly affect individuals.A register of where AI decides, including inside providers.

The seven questions, in CPS 230 terms

Four your CPS 230 work already asks. Three it does not.

1Does a separate CPS 230 nexus exist?Is the party within your group?
2Is there a separate relationship that may engage CPS 230?A service to you, or part of a chain supporting one
3What rights does the entity hold?Passive · significant influence · contractual, including platform, credit and broker agreements · control or board seat
4Where is each party, and does location change what can be obtained?Enforceability, data access and audit rights; offshoring
5What authority has been delegated to the AI?Productivity · analytical · decision · transactional · agentic · actuating
6What could failure do?Immaterial · contract breach · financial loss · customer harm · critical-operation disruption
7What evidence can reasonably be required?None · disclosure · attestation · event notification · audit · continuous telemetry · deterministic control

"No reasonable reach" is a documented conclusion, not a gap. The determination at each step is the entity's and its counsel's.

Where it bites first

A long outsourced chain, no AI team of its own, and a date.

The date might be an APRA review, an incident, a material provider renewal, or the 10 December 2026 disclosure deadline.

Kind of entitySmallMid-sizedLargeVery large
Industry super fund
Administration and mandates outsourced
UnlikelyFirstFirstLater
Customer-owned bank
Core banking run by a provider
UnlikelyFirstEarlyLater
Private health insurer
Claims platforms outsourced
LaterFirstEarlyLater
General insurer writing through intermediaries
Agencies, brokers and aggregators hold the AI
UnlikelyEarlyEarlyLater
Life insurer in a trustee chain
Trustees will ask; AI often built by a foreign parent
UnlikelyEarlyLaterLater
Entity with a large in-house AI team
Short chain; will answer the letter internally first
UnlikelyLaterLaterUnlikely

What moving first gives you

The value is mostly defensive. That is the point.

Sets the provider template

Providers serve several regulated entities. The first to specify the evidence shapes what they build.

Works at counsel's pace

Wrong assumptions are corrected cheaply, before a deadline sets the pace.

Has the reasonable-steps record first

What was asked, found and decided is recorded before APRA or ASIC asks.

Checked by someone independent

The first board to hold the record can say who checked it, and that the checker is independent of audit.

Commonwealth agencies

The same chain question, on agency general counsel's desk.

The policy for the responsible use of AI in government asks each agency for an accountable owner and an impact assessment for every in-scope use, and a register of AI use cases. Automated decisions are disclosed under the Privacy Act from 10 December 2026. The DTA model clauses will carry the evidence terms with providers, and nobody has yet tested what providers can produce against them.

The impact assessment settles what the agency intends. The AI Use Assessment tests what each provider can actually evidence, through the agency's counsel, without redoing the first.

Advisory for agencies and regulated entities →

Regulatory position as at 26 October 2026. Not legal advice: every legal determination is yours and your counsel's.