APRA CPS 220
Risk management framework and risk appetite, which AI risk has to sit within Source→
One AI system, layers of obligation
Every AI system sits inside several rings of rules, from your own policy out to prudential standards.
Your AI policy
Your own rules come first: what AI may do, risk appetite and who approves, applied through the same controls. Recorded as policies and gate rules
Prudential standards
APRA CPS 220 (SPS 220 for super trustees) sets the risk framework and appetite every AI system is rated against. CPS 230 (41 controls) and CPS 234 (24 controls), mapped control by control. Mapped in the control registry
Accountability
The Financial Accountability Regime: key functions (16 controls) tie each AI system to an accountable executive. Mapped in the control registry
Privacy and consumer law
Automated-decision transparency under the Privacy Act starts in December 2026. Referenced
International frameworks
ISO/IEC 42001, NIST AI RMF and the EU AI Act, for groups operating abroad or certifying their AI management system. Referenced
What Australian rules expect of AI
CPS 220 sets the risk framework the rest sits in, three are mapped control by control in the registry, and the rest is recent guidance on duties you already have. Each comes back to the same basics: know your AI, own it, prove it is controlled.
Risk managementCPS 220
Applies since 1 Jul 2019AI risk sits inside the Board-approved risk framework and risk appetite. CPS 220 does not name AI, but it covers the risks AI creates.
What it means for AI- AI risk sits in the framework under the material risk it belongs toEach AI system linked to its material risk
- The risk appetite sets a tolerance for each material riskTier rules set against your Board-approved tolerances
- A yearly Board declaration to APRA, and an independent review every three yearsEvidence kept ready for both
- Significant breaches go to APRA within 10 business daysBreaches recorded with notification dates
SPS 220 sets the same for super trustees, since 1 Jan 2020
Operational riskCPS 230
Applies since 1 Jul 2025AI now sits inside critical operations, and AI vendors are service providers. CPS 230 does not name AI, but it covers it.
What it means for AI- Every AI system in a critical process has an owner and a risk tierInventory with owners, tiers and review dates
- Know which critical operations depend on AI, and the disruption you can tolerateSystems linked to the operations they support
- Plan for an AI system failing or behaving wronglyContinuity controls and incident records per system
- Model providers and AI features in vendor software are service providersVendors and models linked to the systems that use them
41 controls in the registry
Information securityCPS 234
Applies since 1 Jul 2019AI systems hold sensitive data, and agents reach tools through identities and keys, so they sit squarely in scope.
What it means for AI- Track threats specific to AI, such as prompt injection and data leakageThreats catalogued against OWASP and MITRE ATLAS
- Know what data each AI system and vendor model touchesData sensitivity recorded in each risk assessment
- Least privilege for agents, and tests that prove itTool and MCP permissions, enforced at the gate and tested
- AI incidents run on the same 72-hour clock as any otherIncident register with notification dates
24 controls in the registry
Financial Accountability RegimeFAR
Applies since Mar 2024When AI supports a key function, someone named is accountable for it. Banks since March 2024, insurers and super trustees since March 2025.
In CognaOne- Accountable executive and key function recorded for each AI system
- Changes in accountability tracked with dates
- Controls and evidence tied to the accountable person
16 controls for insurance key functions
Letter to industry on AIAPRA
Issued 30 Apr 2026Assurance is not keeping pace with adoption, and boards lean on vendor summaries. APRA will step up supervision.
Expects- An owned inventory of AI tools and use cases
- The AI supply chain mapped, fourth parties included
- People in high-risk decisions, controls on agents
- Assurance before go-live and after
Beware the gap, REP 798ASIC
Published Oct 2024624 AI use cases across 23 licensees. Adoption is outpacing governance, often with third-party models.
Expects- Governance that keeps pace with AI use
- Risk judged from the consumer’s side, bias included
- Accountability kept for third-party AI
- Accurate claims about how AI is used
Automated decisionsPrivacy Act
Applies from 10 Dec 2026Privacy policies must say where a program makes, or substantially helps make, decisions that significantly affect people.
Must disclose- The personal information these programs use
- Decisions made by a program alone
- Decisions a program substantially helps make
General information, not legal advice.
Global standards that shape CognaOne
These standards shape the platform and our research. They move into the registry as they are mapped control by control.
Australia
APRA, ASIC and OAIC- APRA CPS 220Risk management framework and risk appetite, which AI risk has to sit within Source→
- APRA CPS 230Operational risk management, including AI in critical operations and service providers Source→
- APRA CPS 234Information security, including the data and access AI systems hold Source→
- Financial Accountability RegimeA named accountable person for each key function, and the AI inside it Source→
- Privacy Act, APP 1.7 and 1.8Privacy policies explain automated decisions, from 10 Dec 2026 Source→
Australia
APRA, ASIC and OAICAPRA CPS 230
Operational risk management, including AI in critical operations and service providers Source→
APRA CPS 234
Information security, including the data and access AI systems hold Source→
Financial Accountability Regime
A named accountable person for each key function, and the AI inside it Source→
Privacy Act, APP 1.7 and 1.8
Privacy policies explain automated decisions, from 10 Dec 2026 Source→
United States
NIST, MITRE and OWASPEurope
ISO, IEC and the EUAustralia · APRA, ASIC and OAIC
- APRA CPS 220Risk management framework and risk appetite, which AI risk has to sit within Source→
- APRA CPS 230Operational risk management, including AI in critical operations and service providers Source→
- APRA CPS 234Information security, including the data and access AI systems hold Source→
- Financial Accountability RegimeA named accountable person for each key function, and the AI inside it Source→
- Privacy Act, APP 1.7 and 1.8Privacy policies explain automated decisions, from 10 Dec 2026 Source→
United States · NIST, MITRE and OWASP
Europe · ISO, IEC and the EU
Let's connect
Tell us where your AI programme is today. We scope the first engagement to your own AI estate, and every enquiry is handled in confidence.
Contact us