APRA CPS 230 never names AI agents, yet it shapes every AI operations role in an Australian bank, insurer or superannuation fund. It asks regulated entities to identify their critical operations, keep them within set tolerances, control the risks and prove the controls work. The people who review, escalate and record agent decisions are part of that proof.
What does CPS 230 ask of the people who run AI agents?
CPS 230 is APRA's prudential standard on operational risk management. In plain words, it asks a regulated entity to know which of its operations matter most, understand how they could fail, put controls in place, and show that those controls are monitored and tested. It also asks the entity to plan for disruption and to manage the service providers it depends on.
None of that is written for a particular technology. Once an agent drafts claim letters or triages hardship requests, it becomes part of an operation the standard already covers. The people around it become part of the control: the reviewer who approves a held action, the lead who notices a queue backing up, the operator who drops an agent back to read-only when something looks wrong.
For you, that means knowing what the agent may do, acting inside a documented process when it does something unexpected, and leaving a record someone else can check.
Who does CPS 230 apply to?
The standard applies to APRA-regulated entities: authorised deposit-taking institutions, general, life and private health insurers, and superannuation trustees. It does not apply directly to a fintech outside APRA's remit, a telco or a retailer, though contracts with regulated customers often pass similar expectations down the chain.
That matters for your search. If you are looking at AI operations roles in telecoms in Australia, CPS 230 is background rather than a daily rule. If you are applying to a bank's operations team or an insurer's claims function, expect it to come up, sometimes by name and sometimes as "operational resilience" or "operational risk" in the duties.
Material service providers are a second route in. A company that supplies an AI platform or an outsourced review service to a regulated entity may find the entity's obligations written into its contract, including audit rights and incident notification. The wider guide to AI agent jobs and careers explains how these supplier-side roles differ from in-house ones.
What roles and competence does CPS 230 create?
CPS 230 does not create job titles. It creates duties, and employers attach those duties to roles with many names. Read the duties in an advert, words such as review, exceptions, approvals, incident, escalation, control testing and business continuity, rather than hunting for one title.
The roles that tend to carry these duties around agents are:
- Reviewer or case handler. Approves or rejects actions the agent holds for a person, such as a hardship arrangement or a claim settlement letter. Competence means knowing the policy well enough to disagree with the agent.
- Operations or queue lead. Watches volumes, turnaround and exceptions, and decides when the agent's work should be paused or routed to people. This is where tolerance levels for critical operations meet daily work.
- First-line risk or control analyst. Tests that agent controls work, checks samples of decisions and logs findings. The standard expects controls to be tested, so someone has to test them.
- Agent owner. Accountable for one agent's scope, sources and performance, and for what happens when it fails. The mechanics are set out in how CPS 230 applies to AI agents.
The competence that runs through all four is the same: understand what the agent may do, recognise when it is outside its bounds, act inside the incident and escalation process, and leave a clear record.
How do you show employers you meet it?
| Requirement | What employers will expect of you | How to show it |
|---|---|---|
| Identify and maintain critical operations | Know which processes the agent touches and which are critical | Describe one process you worked on, its critical points and who owned them |
| Keep operations within tolerance | Notice when queues, errors or delays breach a set level, and act | An example of escalating a backlog or error trend, with what you recorded |
| Controls that are monitored and tested | Review agent decisions consistently and sample for errors | A record of review decisions to a standard, or a control-testing task |
| Incident management | Follow the incident process, including pausing an agent's actions | Explain a time you stopped work pending investigation |
| Service provider management | Treat the agent platform as a provider with its own controls | Show you can read a provider's control description and ask the right question |
| Clear roles and accountability | Know who decides, and record your decisions under your name | Decision logs, sign-offs and a credential an employer can check |
The table is a translation, not an official mapping. The guide to staff capability under CPS 230 covers what employers build on the training side.
What might an interviewer ask you about CPS 230?
Interviewers rarely want a recital of the standard. They want to know whether you would behave well inside it. Expect questions like these:
- "An agent starts producing letters with the wrong figures on some cases. What do you do in the first hour?" A good answer stops the affected action, keeps the evidence, informs the agent owner and follows the incident process before anyone tries to fix the agent.
- "Which of our operations do you think are critical, and why?" Show that you think about customer harm and the entity's obligations, not only internal convenience.
- "How would you know the agent was getting worse?" Talk about review samples, how often reviewers overturn its drafts, complaint themes and the queue figures a lead watches each day.
- "Who is accountable when the agent gets it wrong?" The entity is, through named people. Your part is to make decisions you can stand behind and record them.
A candidate we will call Mia, moving from a contact centre team lead role to a bank's operations team, answered the first question by describing how she would stop outgoing letters while still allowing drafts, so customers were not left without a response. That level of detail is what interviewers remember.
What does CPS 230 not cover?
CPS 230 is about operational risk. It does not tell you how an agent should handle personal information; that is the Privacy Act 1988, overseen by the OAIC. It does not set conduct rules for selling or advising, which is mainly ASIC's ground. It does not certify individuals, and it does not require any particular course or credential.
It also does not make agents acceptable or unacceptable in themselves. A regulated entity can use agents in critical operations if it manages the risk. The standard asks for evidence, not abstinence. The government's voluntary AI safety standard sits alongside it but does not bind in the same way.
This is an explanation of the rules, not legal advice; confirm how they apply to you with counsel.
Where the Academy fits
The Operator track at Pryme Intelligence Academy covers what CPS 230 asks of front-line people: how a governed agent decides, human review, reading the workspace and what to do when something goes wrong. Certified Pryme Operator includes a two-week practicum of review decisions marked against a standard and signed off by an assessor, so it shows real decisions. It is issued through VerifyDoc.ai and an employer can check it at source.
Questions people ask
Do I need banking experience to get a CPS 230-related role?
No. Many people come from contact centres, insurance claims or operations in other sectors. What helps is evidence that you follow a process under pressure, escalate properly and record decisions. For ways to meet people already doing the work, see how to network into AI operations roles.
Is CPS 230 knowledge useful outside Australia?
Yes, in substance. Other markets have operational resilience rules with similar aims, such as DORA in the EU and the Basel Committee's operational resilience principles. The habits transfer even when the rule names differ.
Should I put CPS 230 on my CV?
Only if you can back it up with an example. A line such as "reviewed agent-held hardship decisions within a CPS 230 control framework, with every decision logged for testing" says more than listing the standard as a skill.