Australian organisations need training in tiers: literacy for everyone, role training for people who use agents, certified competence for people who approve agent actions, and deeper training for the people who own agents. APRA's CPS 230, ASIC's conduct focus and the Privacy Act 1988 shape what each tier must cover. Records of each person's level and currency make it defensible.
What should Australian organisations train staff on?
The core content is the same everywhere agents act on a business's behalf. Staff who work with agents need to understand:
- Depth. What each agent may do, from reading and drafting to executing routine work within bounds or moving money.
- Verdicts. Why an action was allowed, held for a person, refused because it sits outside the agent's bounds, or withheld at the platform floor.
- Provenance. Where each fact came from: sealed by a system of record, signed by a named party, declared by a customer, or arriving from outside. An external claim cannot authorise a payment.
- Lapse. Agent certificates lapse after 90 days, and a run that clears no depth drops an agent to depth 1 the same day. People must expect what an agent can do to change.
- The record. Every verdict is filed, so any decision can be reconstructed, including the person's part in it.
Approvers need more than this: practice on real decisions, an assessment they can fail, and a certificate that is checkable and current.
Which Australian rules shape the training?
Australia does not have one AI training rule. Several frameworks point the same way.
APRA and CPS 230. APRA-regulated entities, such as banks and insurers, must meet CPS 230 on operational risk management. Agents acting in critical operations are part of that risk, and the people overseeing them are a control. Competent oversight, and evidence of it, is part of managing that risk.
ASIC. ASIC regulates conduct in financial services and markets. Where agents draft advice, handle complaints or deal with retail customers, the people approving their work need to be able to stand behind the outcome.
The Privacy Act 1988 and the OAIC. Agents handle personal information, and staff must know what they may disclose and to whom. The note on the Australia Privacy Act and training staff who use agents covers this in depth.
The voluntary AI safety standard. The government has published a voluntary AI safety standard. It is not binding, but it is a reasonable reference for what good practice looks like, including human oversight.
Cross-border rules. Organisations deploying agents in the EU face Article 4 of the EU AI Act, which asks for sufficient AI literacy among staff. ISO/IEC 42001 offers a certifiable AI management system for those who want an external standard.
This is an explanation of the rules, not legal advice; confirm how they apply to you with counsel.
How should training differ by sector?
| Sector | Agent work that reaches people | Training emphasis |
|---|---|---|
| Banking and insurance | Payments, credit, claims, complaints | CPS 230 operational risk, conduct, provenance on payments |
| E-commerce | Refunds, fraud holds, customer replies | Refund evidence, fraud releases, disclosure |
| Logistics | Carrier invoices, claims, exceptions | Contracted rates, claims evidence, shift cover |
| Health and care | Bookings, records, billing | Privacy, the clinical line, identity checks |
Role-specific examples are in the briefs on operations managers in e-commerce in Australia and in logistics in Australia.
An example from a general insurer
A general insurer deploys an agent to answer policyholder enquiries and draft claim payments, with every payment held for a person. Under its CPS 230 framework, the insurer treats claim payments as a critical operation.
It gives every staff member a short literacy module, trains the claims team on the agent's capabilities and refusals, and certifies the 15 people who approve claim payments after a written exam and a supervised practicum. Approval rights depend on a current certificate. When APRA asks how the insurer manages operational risk in claim payments, the insurer can show the agent's depth, the people who approve, their certificates, and the filed record of each payment.
How should you evidence it?
Keep a role map showing who is at which tier, training records with results, certificates checked at source, rights histories, and briefings when agents change. Compare approaches in Spain and Japan for organisations operating across markets. The guide to AI literacy and certification covers the general model.
Where the Academy fits
Pryme Intelligence Academy is global and not tied to one market's rules; its module tests are set in the learner's own industry and role. The free Foundations track covers the literacy tier, the Operator track the approver tier, and Builder and Architect the owner tiers. A free readiness assessment for teams of up to 25 people shows gaps by concept before you commit.
Questions people ask
Does CPS 230 require AI training specifically?
It does not name AI training. It requires sound operational risk management, and where agents act in critical operations, trained oversight is part of meeting it.
Do small businesses outside APRA's scope need this?
They are not bound by CPS 230, but the Privacy Act may apply, and the case for trained approvers holds wherever agents move money or handle customer data.
Is the voluntary AI safety standard worth following?
It is a useful reference for good practice, particularly on human oversight, even though it is not binding.