Superannuation · AI
Guide
How do you get value from AI that your members can rely on?
A secure member-facing assistant and a trustworthy one are different builds. Authentication and encryption decide whether a request was legitimate; they say nothing about whether the answer was right, and it is the answer a fund has to account for. This guide sets out the serving contract and the invocation record that make an AI answer defensible, traces the AI value chain that connects a member's question to the record that explains it later, and names what changes when a significant share of member records sit with an administrator rather than the trustee.
Inside the guide
- 01Why is it hard to find out what good looks like?
- 02What happens when an assistant answers a member?
- 03What is the AI value chain?
- 04What has to be true of your member data?
- 05How do you know the assistant did what you approved?
- 06What do you specify before the first member-facing assistant goes live?
Full guide delivered to your inbox
What the guide covers
What to specify before the first member-facing assistant goes live, and why superannuation cannot wait for the guidance to catch up.
Covers why AI in superannuation is arriving ahead of a settled reference model, what ASIC's 2026 platform trustee review and the Cursor support-bot incident both show about ungoverned answers, the seven-link AI value chain and where trust attaches to it, what a serving contract needs beyond classification and provenance (authority and status), how to build the record of what an assistant actually did, and the four things to specify before the first assistant answers a member.
What's inside
The seven-link AI value chain diagram, a reference architecture showing the AI gateway and the administrator boundary, worked examples on retirement projection assumption sets and successor fund transfers, and a four-point specification checklist for the first assistant deployment.
Get the full guide
Delivered to your inbox. Name and email only.
No noise. Unsubscribe anytime. Your details are used only to deliver this guide and occasional Lumaris insights on the same topic.
24%
Of Australian financial and insurance services businesses used AI in 2024-25, a 24-fold increase from 1 per cent in 2021-22.
Source: Australian Bureau of Statistics, Characteristics of Australian Business, 25 June 2026.
$300bn
In platform trustee member benefits across more than 977,000 member accounts, reviewed by ASIC's 2026 data monitoring assessment that found trustees overwhelmingly disappointing on data use.
Source: Australian Securities and Investments Commission, review of platform trustee data monitoring, June 2026.
10 Dec 2026
Commencement date for Australian Privacy Principle 1.7 to 1.9, requiring disclosure of the personal information used in automated decisions that significantly affect a person's rights or interests.
Source: Privacy and Other Legislation Amendment Act 2024 (Cth), Schedule 1 Part 15.
Martin Barnier
View LinkedIn profileRead enough? The guide is one form away.
Get the guideBefore you download
Questions about this guide
A serving contract is a short, specific statement of what the data platform will hand one named assistant, and what travels with the data: its classification, provenance, permitted purpose, currency, and for an AI consumer specifically, which source is authoritative and whether it is current or superseded. Most funds have never written this down.
Security controls such as authentication, encryption and tenancy decide whether a request was legitimate. They say nothing about whether the answer the assistant gave was correct, and it is the correctness of the answer, not the legitimacy of the request, that a fund has to account for to a regulator or a member.
A significant share of member records sit with an administrator rather than the trustee, so the fund's serving contract is often describing data it does not hold. CPS 230 already classifies fund administration as a critical operation and material service provider for most RSE licensees, so the specification has to reach across that boundary as a contract term, not just an architecture decision.
Four things: a serving contract per named assistant naming the authoritative and current sources, a record of what happened at every invocation, a named and owned corpus narrower than whatever the assistant's underlying permissions would allow, and proof that any answer a member could act on can be traced back to its sources.
Let us talk
If the guide surfaces something you want to work through, that is a good place to start.
Most conversations begin simply. Someone wants to know whether we are the right fit for what they are navigating. That is a perfectly good starting point.