APRA CPS 234 Penetration Testing Requirements (2026 Guide)

Here is the position every organization bound by APRA CPS 234 is in. APRA can hold your Board personally accountable and demand independent assurance — if you can't show at least annual, independently-conducted testing of your security controls (with results escalated on unremediated gaps), you're exposed to a breach notification, an APRA enforcement action, and a failed audit.

Most teams answer this the expensive way: they buy a single, five-figure, point-in-time manual penetration test, sweat through weeks of scheduling, then hope nothing changes until the next cycle. It always changes. This guide breaks down exactly what the APRA CPS 234 penetration testing requirements demand, what an assessor needs to see, and how to stay continuously in-scope — without a five-figure invoice every time your environment moves.

CPS 234 paragraphs 27–28 mandate a "systematic testing program" to test the effectiveness of information security controls, with frequency/nature commensurate with rate of threat/vuln change, criticality of assets, and materiality of change — at a minimum annually and whenever there is a material change to information assets or the business environment, conducted by "appropriately skilled and functionally independent specialists." CPS 234 is principles-based and does not use the literal words "penetration test," but its guidance CPG 234 explicitly names penetration testing (network protection, unauthorised access, software security) and red-team testing as the expected methods, making a periodic pentest the de-facto evidence of compliance. Deficiencies that can't be remediated timely must be escalated to the Board/senior management.

Who must comply: All APRA-regulated entities in Australia: authorised deposit-taking institutions (banks, neobanks, credit unions/building societies), general insurers, life insurers, private health insurers (PHIPS Act), RSE licensees / superannuation funds, and non-operating holding companies. Obligations extend contractually to third parties/service providers that manage the entity's information assets.

A compliant test has to hit the environment from two directions — the external internet-facing surface and the internal surface reachable from inside your network — and at both the network and application layers. The most common reason a first attempt fails review is incomplete scope: the public site was tested, but the internal APIs, admin panels, and the controls that isolate sensitive systems were not.

A traditional manual penetration test runs roughly $10,000–$30,000 and lands as a PDF that is accurate for about a day. Then your environment changes and you are out of coverage until the next cycle. That is a lot of money for a snapshot, and it leaves you blind between assessments.

Pentestas is pentesting as a service: an AI penetration testing system that runs APRA CPS 234-aligned internal and external tests on your environment continuously — automatically, after every deploy — starting at $79/month. It is penetration testing with AI at the core, orchestrated by frontier models (we run penetration testing with Claude for deep reasoning and cost-controlled penetration testing with DeepSeek for high-volume passes), delivered as a B2B SaaS pentest platform that maps every finding straight to the requirement it satisfies.

The platform exercises the external surface from the internet and the internal surface through a lightweight agent inside your network, so one engagement covers both directions. It tests the application and network layers, re-tests every exploitable finding after you fix it (capturing before-and-after evidence), and — because it re-runs after every change — handles the “after significant change” obligation automatically instead of turning it into a fire drill. Every finding ships with real exploit evidence, a severity rating, and remediation steps.

Bottom line: APRA CPS 234 is not satisfied by one lucky PDF per cycle. It rewards continuous coverage of the whole environment, internal and external, re-tested after every change — exactly the shape of problem an AI penetration testing system is built to solve, at a fraction of the cost of being compliant for one day out of the year.