What does APRA AI Letter to Industry (April 2026) say about AI in banking?

What are the key points of APRA AI Letter to Industry (April 2026)?

• Evidence base: a targeted engagement on selected large banks, insurers and superannuation trustees in late 2025; the letter's attachment is an 'AI Supervisory Engagement Debrief: Observations for Executive Management'.
• Boards: APRA observed many boards still developing the technical literacy to challenge AI risk and an overreliance on vendor presentations and summaries; it expects boards at a minimum to maintain sufficient AI literacy and oversee an AI strategy consistent with risk appetite and tolerance, with monitoring and reporting including third-party dependencies and clear triggers.
• Information security: attack pathways include prompt injection, data leakage, insecure integrations, exploit injection and manipulation of autonomous AI agents; identity and access management has not yet adjusted to non-human actors, and AI-assisted development strains change and release controls.
• Governance: few entities have operationalised governance for AI; APRA expects frameworks and reporting lines, ownership across the AI lifecycle, an inventory of AI tooling and use cases, human involvement for high-risk decisions, and staff training.
• Suppliers: some entities depend heavily on a single provider for multiple AI use cases and few had tested exit and substitution; APRA expects mapping of the full AI supply chain including third and fourth parties, contractual transparency and auditability, and active concentration risk management.
• Assurance: point-in-time and sample-based methods are ill suited to probabilistic models that learn, adapt and degrade; APRA expects integrated assurance, technically capable second-line and internal audit functions, and continuous monitoring proportionate to criticality.
• Supervision: APRA is finalising a forward plan for AI supervision covering entity prudential reviews, thematic activities and AI supplier engagement, and says it will consider whether further policy action is needed.
• Enforcement: where entities fail to identify, manage or control AI risks proportionately to size, scale and complexity, 'we will take stronger supervisory action and, where appropriate, pursue enforcement'; the letter was signed by APRA Member Therese McCarthy Hockey.

What did APRA AI Letter to Industry (April 2026) change for banks?

It introduced no new standard but converted existing technology-neutral requirements (chiefly CPS 230 on service providers and operational resilience and CPS 234 on information security) into specific supervisory expectations for AI, and put banks on notice that AI supervision, including engagement with AI suppliers, is being built. It preceded APRA and ASIC's joint frontier AI roundtables, summarised in a media release on 27 August 2026.

What does APRA expect of banks managing AI risk?

APRA's 30 April 2026 letter expects banks to manage AI risk within existing prudential standards across five areas. Boards must hold enough AI literacy to challenge and oversee an AI strategy that fits risk appetite and covers third-party dependencies. Information security must address AI-specific attack paths, with privileged access management, timely patching, security testing of AI-generated code and controls over agentic workflows. Governance must include frameworks, lifecycle ownership, an inventory of AI tools and use cases, human involvement in high-risk decisions and staff training. Supplier risk management must map the whole AI supply chain including fourth parties, secure contractual transparency and manage concentration and exit. Assurance must be integrated and continuous, with second-line and internal audit capable of assessing probabilistic and agentic systems. APRA says it will take stronger supervisory action and, where appropriate, pursue enforcement.

The letter is APRA's way of saying AI does not need a new standard to be supervised. Most entities, APRA found, recognise that existing prudential standards apply to AI risk but few have operationalised that, and many treat AI as 'just another technology', missing the distinct characteristics of predictive systems, adaptive model behaviour, bias, and privacy and data risks. Weak controls were found over post-deployment monitoring, model behaviour monitoring, change management and decommissioning of AI capabilities.

APRA also states that it is engaging across the sector on the potential for increased cyber threats from high-capability frontier AI models, has been engaged with the Council of Financial Regulators and government agencies, and refers entities to current Australian Signals Directorate advice on frontier models. That thread continued in the joint APRA-ASIC statement of 27 August 2026, which reports nine industry roundtables in June and July 2026 attended by more than 600 people.

ASIC's REP 798 covers the conduct and consumer side of the same governance-gap theme. APRA's letter is the prudential side, concentrated on boards, executives and cyber and supplier resilience.

WHAT THIS MEANS IN PRACTICE

• Brief the board with an AI-risk pack covering the strategy, risk-appetite alignment, third-party dependency reporting and triggers; avoid relying on vendor presentations alone.
• Build and maintain an inventory of AI tooling and use cases, with named owners across design, deployment, monitoring and decommissioning.
• Replace policy-only controls on staff use of enterprise AI tools with enforceable technical restrictions and preventative controls.
• Map AI supply chains to fourth parties, test exit and substitution for critical AI providers, and review contracts for audit rights, model updates, incident notification and data-handling changes.
• Upskill second-line risk and internal audit so they can assess probabilistic models and agentic workflows, and move from point-in-time sampling to continuous validation.