APRA
APRA prudential standards and AI: CPS 230, CPS 234, and CPS 220 translated into practical compliance for Australian financial services.
56 articles
Articles about APRA
Your AI Vendor Count Is Hiding One Foundation Model.
Three contracted AI vendors can still fail as one service when they share a foundation model, cloud control plane or identity layer. Map every service to its common technical roots, then test whether your nominated substitute survives the same failure.
Read article
AI Can Draft the APRA Warning Comment. It Cannot Sign Off the Return.
APRA Connect may accept an explanatory comment after a validation warning. That does not make the explanation true. Build every AI-drafted sentence from the return cell, source query, data owner, validation rule and named reviewer who accepts responsibility for submission.
Read article
Your GenAI Approval Needs an Expiry Trigger.
A GenAI approval records what passed on one date. It does not prove the live system still deserves approval. Build a living validation passport with explicit failure modes, change triggers, expiry and a human-controlled stop-use rule, and keep the four regulatory layers behind it separate.
Read article
Defensible AI Lineage Starts Before the Prompt.
A citation beside an AI answer is not data lineage. You need the source snapshot, transformations, retrieved context, prompt, available model configuration, raw output and human edits. Without that packet, assurance starts after the evidence trail has already broken.
Read article
Skill Files Are Controlled Documents. Treat Them Like It
The moment a skill file shapes work a regulator cares about, it stops being a personal productivity hack and becomes a controlled document. Version numbers and change logs inside the file, a named owner and approver, retained test evidence, reviews triggered by obligation changes, and a register of approved skills. Here is the discipline, anchored by a board-reporting skill your risk team can adapt today.
Read article
Every Model You Rely On Has a Retirement Date
Claude Opus 4.1 stops answering on 5 August 2026. OpenAI shut down sixteen model snapshots on 23 July 2026 and will retire its Agent Builder and Evals platform on 30 November 2026. These dates are published, they are not negotiable, and the notice you receive depends on a product tier most teams never consciously chose.
Read article
A Skill File for Regulatory Change: Impact Assessment On Demand
Every regulator release triggers the same scramble: read it, work out what it touches, brief the owners. The release changes every time. The method does not, and a stable method is exactly what a skill file is for. Here is a complete regulatory change impact assessment skill, the five prompts that build your team's own version, and how to install it in ChatGPT and Claude.
Read article
Fine-Tuning Writes Your Data Into the Model
Every organisation customising AI faces the same build decision: prompt it, retrieve for it, or fine-tune it. It is usually argued on accuracy and cost. The axis nobody raises is reversibility. Retrieval leaves your data in a store you own, audit and delete. Fine-tuning copies it into a model artefact you cannot meaningfully un-write, which makes the customisation call a records decision before it is an engineering one.
Read article
Internal Audit's AI Say-Do Gap Now Has Numbers
Two 2026 surveys put hard numbers on something internal audit has felt for a year: adoption is racing ahead of capability. Gartner has 83 per cent of audit functions piloting or using AI, while only a small share feel able to embed it. The IIA and AuditBoard have 85 per cent alert to AI-enabled fraud and fewer than 40 per cent ready to detect it. For an Australian function, the gap is a benchmark to self-assess against, not a headline to read past.
Read article
Canberra's AI Register Mandate Is a Preview
The Australian Government has made an AI use-case register mandatory for its own agencies. For private-sector GRC teams building registers under CPS 230, the government template is a working preview worth studying now.
Read article
FAR Eased Up. Your AI Map Still Holds
ASIC and APRA are trimming FAR reporting from 16 June 2026. The accountability behind your AI use-case register has not moved. The practical change is what you actively maintain versus what you only produce on request.
Read article
Your AI Logs the Tokens. Not the Decision.
Your AI telemetry records what the call cost, not what the model was told or what it said. Every content attribute in the OpenTelemetry GenAI conventions ships opt-in, and the conventions warn the message content is likely to hold personal information. That is a privacy default, not an audit default, and the two pull in opposite directions. Here is the capture, redaction and retention decision nobody is being asked to make.
Read article
Super Trustees, AI and the Discretion You Cannot Delegate
APRA has put superannuation trustees inside its AI expectations and wants human involvement for high-risk decisions. The harder limit is statutory. A trustee's core discretions are owed personally to members under the SIS Act, and the money spent buying the AI has to clear the best financial interests duty. Here is how to tell a decision AI can support from one it must never make.
Read article
Your AI Agent Can Remember Now. Govern What It Keeps.
In 2026 the major labs shipped persistent memory as a first-class agent feature, barely six weeks apart. An agent that remembers across sessions is a different thing to govern, and a new attack surface. Treat the memory store as a governed data asset with write rules, provenance, expiry and rollback, not invisible plumbing.
Read article
Freed Board Bandwidth Is for AI Oversight
APRA's draft CPS 510, released 16 June 2026, consolidates five governance standards and cuts duplicative fit-and-proper paperwork for around 6,000 people. Read with APRA's April AI letter, the freed board capacity has an obvious destination: AI oversight.
Read article
Your AI Passed the Pilot. Production Is a Different Test.
A pilot that passed proves the AI worked once, on that data, on that version of the model. Then the model changes underneath you, the data shifts and usage drifts, and the thing you signed off quietly stops being the thing running. The answer is a standing evaluation harness, not a launch-day test. Here is how to build one, and why regulated work needs it most.
Read article
Australia Will Not Pass an AI Act. You Are Still Regulated.
The National AI Plan settled the question every GRC team was waiting on. Australia will not pass a standalone AI Act. That is not a reprieve. It means AI is already regulated, spread across the laws and regulators you answer to now. Here is how to stop waiting for an AI law and map every AI use to the obligation it already touches.
Read article
More Agents Is Not More Intelligence. Govern the Coordination.
The plumbing for multi-agent AI just got standardised, so building systems where agents talk to each other is now easy. That is exactly why the useful question changed. Not can you wire agents together, but should you, and how do you govern the coordination when more agents is not more intelligence.
Read article
The 2026 AI Governance Talent Market: Assurance Skills Move to the Core
Australia's AI governance market is shifting from principle-setting towards assurance: control design, model testing, data lineage, third-party oversight and evidence a board or auditor can challenge. APRA has named the skills gap, no salary guide prices the role cleanly, and the scarce profile is the practitioner who can move from principle to proof.
Read article
Model Routing Cuts AI Bills. It Also Moves Your Data.
The enterprise AI story has shifted from which model is best to which one you can afford to keep using, and buyers are moving from tokenmaxxing to model routing. The instinct is right. But for Australian regulated work a cheaper model is usually a different provider in a different place, so every routing rule is also a Privacy Act and APRA data-flow decision.
Read article
Someone Poisoned the Tool Description. The Agent Did the Rest.
Microsoft's incident-response team has documented the first real-world attack on the Model Context Protocol: poison a tool's description and you redirect the agent, without touching its code, its credentials or its prompt. The site's earlier MCP piece predicted this. Here is what "least agency, not just least privilege" means as an operational control, not a slogan.
Read article
AI Guardrails: The Safety Layer No Vendor Can Ship for You
A new default model does not make your AI system safe. Reliability and safety in production come from the guardrail layer you build around the model: input rails, grounding, output rails and action gates. Here is the architecture, a worked example that turns a written policy into running rails, and the prompts and checklist to build your first rail this week.
Read article
You Are Now Buying Software for Agents, Not People
Gartner says $234 billion of enterprise SaaS spend is exposed to agentic arbitrage by 2030. Read as a vendor problem it is a headline. Read as a buyer it changes how you procure and govern the software you already run, so this piece turns the forecast into an API-parity test you can run this week, a five-clause contract checklist and a Monday workflow.
Read article
Claude Sonnet 5 Became the Default. That Is a Change Event.
Anthropic released Claude Sonnet 5 on 30 June and made it the default in Claude Code and on Claude.ai Free and Pro. Almost nobody chooses a model, so a frontier swap is a silent change to a system you may have already validated. Here is how to treat it as a change event: the prompts, the Monday workflow and the register entry.
Read article
Fable 5 Returns With a Jailbreak Severity Framework
Claude Fable 5 comes back globally on 1 July, three weeks after a US directive pulled it. The return matters less than what came with it: a safety patch that over-blocks routine coding, and a four-dimension jailbreak-severity framework the major labs are building together.
Read article
CPS 230's 1 July Deadline Just Caught Up With Your AI Vendors
From 1 July 2026, pre-existing contracts with material service providers must meet APRA's CPS 230, and a growing share of those arrangements are now AI. Here is the work to do before the deadline, plus a reusable contract-review prompt.
Read article
AI Cyber Defence Just Scaled Up. Mind Your Open-Source Dependencies.
OpenAI pointed its most capable cyber model at the open-source software the world runs on, and found hundreds of real flaws in days. The capability is dual-use. Here is what it means for Australian teams.
Read article
AI Agents Just Went From Minutes to Hours. The Control Point Is Where They Run.
OpenAI bought Ona and published research showing AI agents now run for hours, not seconds. The unit you have to govern moved from the prompt to the environment the agent runs in.
Read article
Your AI Assistant Just Became a Shared Teammate. Govern the Channel.
On 23 June, Anthropic launched Claude Tag, a single shared Claude that lives in a Slack workspace with its own memory and admin-scoped access to channels, tools and data. The unit of AI collaboration just moved from the private conversation to the team channel. The thing you now have to govern is no longer a prompt. It is a standing presence. Here is what changes, and the three decisions to make before it is live.
Read article
Model Context Protocol: The Standard Wiring AI Into Your Tools
In eighteen months the Model Context Protocol went from an Anthropic experiment to the way AI plugs into your tools and data. Understanding what it is matters less than governing the connectors, because each one is a new door into your systems. Here is the capability and the control work.
Read article
The Strongest Open Model Is Now Chinese. Mind Where Your Data Goes.
On 16 June, China's Z.ai released GLM-5.2 under an MIT licence with no regional limits, the highest-ranked open-weights model on its own coding benchmarks. With Anthropic's Fable 5 pulled by a US directive, the strongest model you can simply download and run is now Chinese. The decision that carries your risk is not the model. It is whether you run the open weights yourself or send your data to the hosted API.
Read article
Business Teams Can Now Build Their Own AI Agents
Databricks launched Genie One this week, an agentic coworker pitched at finance and marketing teams, not engineers. The real shift is who holds the build button, and where that moves the control point. Here is what to do this week, with a governance prompt you can run today.
Read article
The OWASP Agentic Top 10: A Defence Playbook for the Agents You Are Deploying
OWASP has published a Top 10 built specifically for AI agents. It reframes the agent as a privileged user that reads untrusted text and acts with your access. Here is the practical defence playbook.
Read article
AI Is Moving Into the Core Systems of Regulated Work
This week two of the world's largest IT services firms began wiring a frontier model into the core systems that banks, insurers and airlines run on, not the chat window. Here is what it means for regulated work, and what to do this week.
Read article
AI Week in Review, 8-14 June 2026: A Frontier Model Pulled by Government Order
The week a US directive forced Anthropic to suspend two new frontier models worldwide, plus six verified vendor moves and a repeatable method for turning AI news into Monday actions.
Read article
Build the Knowledge Spine That Stops Generic AI Output
Generic AI output is a context problem, not a prompt problem. Learn how a governed knowledge spine grounds your models in real organisational knowledge, with worked examples for GRC, workers compensation and HR.
Read article
Build an Offline GRC Controls Console Without Creating Shadow IT
A single-file controls console can sharpen evidence review without leaking data. Treat it as a governed end-user computing tool, not a free win.
Read article
Claude Fable 5: Frontier Capability, With Conditions Attached
Anthropic has put a Mythos-class model on general release, and the conditions matter as much as the capability. A silent classifier fallback, a mandatory 30-day retention policy and a 23 June billing switch all belong in your next third-party AI assessment.
Read article
Microsoft's Seven MAI Models: The In-House Bet Under Copilot
Microsoft launched seven home-grown MAI models at Build 2026 and started swapping them into Copilot and the Microsoft 365 stack. For practitioners the story is procurement, not benchmarks: data lineage claims, weight tuning, and a billing change in the same week.
Read article
AI in Internal Audit: What Still Counts as Evidence
Internal audit functions are adopting AI faster than they are writing the rules for their own use of it. The IIA's 2024 Standards never mention artificial intelligence, yet every evidence, documentation and objectivity requirement still applies to AI-assisted audit work.
Read article
Build an AI Use Case Register That Boards Can Actually Use
Practical guidance for GRC teams to create AI use case registers that deliver clear, decision-ready evidence for boards and risk committees.
Read article
AI Cyber Risk Is Now a Board Governance Issue
ASIC's May 2026 cyber uplift warning highlights that AI-driven cyber risk demands active board and risk committee oversight, not just IT fixes. This article outlines a practical governance operating model for GRC teams.
Read article
AI Agents Need Approval Gates Before They Need Autonomy
Autonomous AI agents are becoming practical, but organisations should design approval gates, permissions and evidence trails before granting action rights.
Read article
From Voluntary AI Guardrails to Audit Evidence
Australia's voluntary AI guardrails only become useful when GRC teams translate them into control objectives, artefacts and assurance tests.
Read article
Small Models, Edge AI and the Next Governance Blind Spot
As AI moves into devices, business apps and smaller specialised models, organisations need governance that looks beyond frontier models and public chatbots.
Read article
AI Incident Response Needs an Evidence Pack, Not Just a Playbook
Prompt injection, data leakage and agentic failures require GRC teams to rethink incident response evidence, escalation and assurance.
Read article
Board AI Literacy Is Now a Control Expectation, Not a Training Nice-to-Have
APRA's April 2026 AI letter signals that board AI literacy is becoming a governance control expectation, not a generic awareness exercise.
Read article
ASIC's AI Supervisory Posture, Decoded
ASIC's posture on AI in financial services is now visible across REP 798, the 2026 Key Issues Outlook, and recent statements from the Chair. Five themes shape supervisory expectation, and three create immediate work for compliance teams.
Read article
CPS 234 and AI Vendors: A Due Diligence Framework
CPS 234 has been in force since 2019. AI vendors stretch the framework in specific ways: training data exposure, model update opacity, and inference infrastructure that crosses the standard's information asset boundaries. A practical due diligence framework.
Read article
FAR and AI: How Accountability Maps to Tooling Decisions
The Financial Accountability Regime makes specific senior executives answerable for the systems and decisions inside their portfolios. AI tooling decisions sit inside that accountability, whether they are formally documented in the accountability map or not.
Read article
Privacy-Safe AI for Regulated Work: A Working Practitioner's Guide
Workers compensation, GRC, HR and clinical roles all sit on regulated data. Using AI well in those roles is not optional. Doing it safely is not optional either. This is the practitioner's guide.
Read article
RAG Explained for Non-Engineers: How AI Reads Your Documents
RAG is the architecture behind most enterprise AI tools you will meet in 2026. The acronym hides a simple idea. Here it is, explained in plain English with a working analogy.
Read article
APRA's Model Risk Thematic Review: What to Expect
APRA's AI supervisory work has moved from signal to substance. The April 2026 letter to industry reported a deep-dive on the largest banks, insurers and super trustees, and a forward plan of prudential reviews and thematic activities is being finalised. Where the pressure lands, and the work to do now.
Read article
CPS 230 and AI: A Practical Operational Resilience Playbook
CPS 230 has been live since 1 July 2025, and APRA's amended standard commenced on 1 July 2026. The question for boards and operational risk teams is no longer whether AI tools fall inside the standard. It is how to evidence it.
Read article
Australian AI Safety Standard: 18-Month Review
Eighteen months in, Australia's voluntary AI Safety Standard has shifted from optional reading to procurement table stakes. Three things worked. Two did not. The next phase is moving towards mandatory.
Read article
GPT-5 in the Enterprise: 60-Day Debrief
Sixty days after GPT-5 hit enterprise GA, the tool-use story is real and the pricing story is messier. Three patterns separate the teams getting value from the teams burning credits.
Read article