> ## Content Index > Fetch the complete content index at: https://kinnatic-informer.ghost.io/llms.txt > Use this file to discover other available public pages before exploring further. # APRA's AI Letter in Practice: From Vendor Decks to Evidence (Inventory, Fallbacks, Assurance) - URL: https://kinnatic-informer.ghost.io/apras-ai-letter-in-practice-from-vendor-decks-to-evidence-inventory-fallbacks-assurance/ - Published: 2026-10-15T16:44:19.000Z - Updated: 2026-10-16T08:17:01.000Z - Author: Marty Przybylak Five months ago APRA wrote to all APRA-regulated entities about AI. The question now is whether boards have changed what they ask to see. One line in it should sting anyone who sits on a risk committee. APRA noted "an over-reliance on vendor presentations and summaries without sufficient examination of key AI risks such as unpredictable model behaviour and the impact on critical operations." In plain English: too many boards are being briefed on AI by the people selling it. This piece turns the letter's themes into the evidence a board should be able to open, test, and challenge, across three words the letter keeps returning to: **inventory**, **fallbacks**, and **assurance**. It is **not legal advice**. Read the letter in full, and confirm how it applies to your entity with your advisors. ## What the letter actually says (short version) APRA ran a targeted engagement with selected large banks, insurers and superannuation trustees in late 2025, then published what it found. The headline: governance, risk management, assurance and operational resilience are "not keeping pace with AI adoption." For boards, APRA set a minimum. Maintain enough AI literacy "to set strategic direction and provide effective challenge and oversight." Oversee an AI strategy consistent with risk appetite, supported by monitoring and reporting, including for third-party dependencies, "with clearly defined triggers aligned to resilience objectives." For executives, the attachment is blunt. APRA saw a tendency to treat AI risk as "just another technology." Staff use enterprise AI tools outside approved control frameworks, and in many cases entities relied primarily on "policy direction or detective, after-the-fact measures" rather than "enforceable technical restrictions or robust preventative controls." Identity and access management capabilities "have not yet adjusted to nonhuman actors such as AI agents." And the consequence: where entities fail to adequately identify, manage or control AI risks, APRA "will take stronger supervisory action and, where appropriate, pursue enforcement." It also said it is finalising a forward plan for AI supervision, covering prudential reviews, thematic activities and AI supplier engagement. That is not a call for a new framework document. It is a call for proof. ## 1\. Inventory: you can't challenge what you can't list **What APRA asked for.** Among its minimum governance expectations: "an inventory of AI tooling and AI use cases," with ownership and accountability across the lifecycle, "from design and development through to deployment, monitoring and decommissioning." On suppliers, it asked entities to map "the full AI supply chain, including material, third‑party and fourth‑party dependencies." **What the vendor deck shows.** The approved pilots. A roadmap. A logo slide of partners. **What evidence looks like.** - A living register of AI tools and use cases: built in-house, bought, and embedded in platforms you already pay for. - Sanctioned and unsanctioned use, side by side. APRA specifically flagged staff using enterprise AI tools outside approved frameworks. - A named owner and lifecycle status for every entry, including what has been switched off. - Upstream dependencies: which models and providers sit underneath, and where one provider supports several use cases. APRA observed "some entities heavily dependent on a single provider for multiple AI use cases." - AI agents treated as identities, with their access recorded like any other privileged actor. The test a director can apply: ask for the register, pick one customer-facing use case at random, and ask who owns it, which model sits underneath it, and when it was last reviewed. If the answer takes a week, the inventory is an intention. ## 2\. Fallbacks: what happens when the AI stops, drifts, or misbehaves **What APRA asked for.** "Where AI supports critical operations, credible fallback processes are required." On suppliers, APRA observed that few entities had "demonstrated robust contingency planning or tested exit and substitution strategies for critical AI providers," and contracts showed "limited evidence of specific provisions addressing audit rights, model updates and deviations, incident notification or changes to data handling." **What the vendor deck shows.** Uptime figures. A resilience slide. A promise that the model is "enterprise-grade." **What evidence looks like.** - A list of the critical operations AI now touches, mapped to the fallback for each: manual process, alternative provider, or degraded mode. - Evidence the fallback has been **tested**, not just documented: when, by whom, and what broke. - The board's "clearly defined triggers": the thresholds at which a use case is paused, rolled back, or escalated, and who has authority to pull them. - Contract clauses you can point to on audit rights, model change notification, incident notification, and data handling changes. - A credible exit or substitution path for each critical AI provider. The test: "If this provider changed its model overnight, or went down for a day, what would we do, and when did we last prove it?" ## 3\. Assurance: point-in-time sampling won't catch a model that changes **What APRA asked for.** APRA observed "reliance on point in time and sample based assurance methods, despite these methods being ill suited to probabilistic models that learn, adapt and degrade over time." Few entities had continuous validation or monitoring to detect "model drift, bias, failure modes, or control breakdowns in a timely manner." It also noted that many internal audit and risk functions "lack the specialist skills and tools required to engage in AI assessment or audit." Its expectations include integrated assurance across cyber, data governance, model performance, operational resilience, privacy and conduct risk, and monitoring that is "continuous and proportionate to the criticality of the use case." **What the vendor deck shows.** A certification logo. A summary of the vendor's own testing. **What evidence looks like.** - Monitoring outputs over time, not a single test report: drift, error rates, overrides, complaints. - A trail that reconstructs what ran, under which **AI Policy** version, by which identity, human or agent. - Second-line and internal audit with the access and tooling to test AI independently, rather than relying on the vendor's summary. - Preventative controls that sit between people (and agents) and the model, not only a policy PDF and an after-the-fact report. The test: "Show me last month's monitoring for our highest-risk AI use case, and the last time it triggered an action." ## Why this matters this quarter Three things since April have raised the stakes. **ASIC is starting its own review.** ASIC's 2026–27 supervisory letter to banks, published 30 September, says it will commence a banking sector AI review "focused on new and proposed use cases of AI by banks and the impact on customers," with expected commencement in Q2 2026–27\. ASIC says it will collaborate with APRA to minimise duplication, and wants consumer protections maintained "particularly for customer-facing interactions, decision-making and lending processes." The same inventory answers both regulators. **The OAIC has published its how-to for automated decisions.** On 30 September the OAIC updated Chapter 1 of the APP Guidelines for the APP 1.7–1.9 transparency obligation, which starts **10 December 2026**. Privacy policies will need to describe where computer programs make, or substantially and directly inform, decisions that could significantly affect individuals. The OAIC says a decision can be in scope "even where a computer program output does not replace the entire decision-making process or is subject to human review." You can't write that disclosure without the inventory. **Parliament is asking the assurance question out loud.** The Joint Select Committee on AI held public hearings last week, with OpenAI, Anthropic, Microsoft and Google appearing alongside banks, unions and employer groups. Ahead of the hearings, chair Jo Briskey MP said, as reported by the Guardian: "I don't think we get the benefit unless we can assure and mitigate against the risks." Three different forums, one operating ask. **Evidence, not intentions.** ## A board evidence checklist A discussion guide for the next risk committee. Adapt it to your entity's size, scale and complexity. ### Discover: what AI is actually in use? - Register of AI tools and use cases, including embedded vendor AI and unsanctioned use. - Named owner and lifecycle status for each. - Third- and fourth-party dependencies, and single-provider concentration. ### Govern: who decides, and what stops it? - Risk tier and human involvement for high-risk decisions. - Board-approved triggers to pause, roll back, or escalate. - Enforceable **AI Policy** controls, not only handbook direction. - Contract clauses on audit, model change, incident notification and data handling. ### Assure: can we prove it works? - Tested fallbacks for AI supporting critical operations. - Continuous monitoring proportionate to criticality. - A reconstructable trail an auditor or supervisor can inspect. - Second line and internal audit able to test independently. If several answers are "the vendor says so," you have a discovery gap, not a reporting gap. ## Discover → Govern → Assure APRA's letter is principle-based and, as it says itself, "technology and vendor agnostic." It doesn't tell you which tool to buy. It tells you what you need to be able to show. The defensible sequence is the same one behind every regulator ask this quarter: 1. **Discover**: see what AI is actually in use. 2. **Govern**: enforceable **AI Policy** and named ownership before AI acts. 3. **Assure**: a trail directors, auditors and supervisors can inspect. That is how Kinnatic is built: an AI governance platform for regulated entities. ## Start with visibility The first evidence a board should ask for is the inventory. Fallbacks, triggers and monitoring all hang off it. Kinnatic.ai can help: surface which AI tools show up in your organisation, compare that with what your register says, and walk into your next risk committee with a gap list instead of a vendor summary. Built for teams who need evidence, not intentions. --- **Disclaimer:** This article is for general educational purposes only and is **not legal, regulatory, or compliance advice**. APRA's letter sets out observations and expectations for APRA-regulated entities; read it in full and confirm how it applies to you with your advisors and supervisor. ASIC's review timing is "expected" per its supervisory letter and may change. Committee hearings and the OAIC guidance are referenced as published at the time of writing; verify against primary sources before acting. ### Sources (primary unless noted) - APRA, Letter to Industry on Artificial Intelligence (AI), 30 Apr 2026: [https://www.apra.gov.au/news-and-publications/apra-letter-industry-artificial-intelligence-ai](https://www.apra.gov.au/news-and-publications/apra-letter-industry-artificial-intelligence-ai?ref=kinnatic-informer.ghost.io) - ASIC, Supervisory letter: banking sector priorities 2026–27 (26-232MR), 30 Sep 2026: [https://download.asic.gov.au/media/0tobhg1c/26-232mr-asic-supervisory-letter-banking-published-30-september-2026.pdf](https://download.asic.gov.au/media/0tobhg1c/26-232mr-asic-supervisory-letter-banking-published-30-september-2026.pdf?ref=kinnatic-informer.ghost.io) - OAIC, APP Guidelines Chapter 1 (updated 30 Sep 2026; APP 1.7–1.9 from 10 Dec 2026): [https://www.oaic.gov.au/privacy/australian-privacy-principles/australian-privacy-principles-guidelines/chapter-1-app-1-open-and-transparent-management-of-personal-information](https://www.oaic.gov.au/privacy/australian-privacy-principles/australian-privacy-principles-guidelines/chapter-1-app-1-open-and-transparent-management-of-personal-information?ref=kinnatic-informer.ghost.io) - Parliament of Australia, Joint Select Committee on Artificial Intelligence: [https://www.aph.gov.au/Parliamentary\_Business/Committees/Joint/Artificial\_Intelligence](https://www.aph.gov.au/Parliamentary%5FBusiness/Committees/Joint/Artificial%5FIntelligence?ref=kinnatic-informer.ghost.io) - The Guardian (secondary), hearings preview and chair's comments, 5 Oct 2026 (Europe) / 6 Oct AEST: [https://www.theguardian.com/australia-news/2026/oct/06/openai-must-explain-action-taken-to-stop-ai-hacking-australians-private-data-chair-of-federal-inquiry-says](https://www.theguardian.com/australia-news/2026/oct/06/openai-must-explain-action-taken-to-stop-ai-hacking-australians-private-data-chair-of-federal-inquiry-says?ref=kinnatic-informer.ghost.io) - Capital Brief (secondary), ASIC makes AI a supervisory priority for 2026–27, 30 Sep 2026: [https://www.capitalbrief.com/briefing/asic-makes-ai-a-supervisory-priority-for-2026-27-8a0e8327-edfc-4791-a39d-e6cf10538a55/](https://www.capitalbrief.com/briefing/asic-makes-ai-a-supervisory-priority-for-2026-27-8a0e8327-edfc-4791-a39d-e6cf10538a55/?ref=kinnatic-informer.ghost.io)