APRA letter to industry on artificial intelligence

What it is

A letter from APRA to every entity it regulates, dated 30 April 2026, reporting what it found when it looked at how banks, insurers and superannuation trustees are adopting AI, and setting out what it expects them to do about it. It is not a new prudential standard. APRA says its framework is technology and vendor agnostic and that the existing standards already apply to AI risk. The letter tells regulated entities how APRA will read those standards when AI is involved, and warns that it will take stronger supervisory action, and enforcement where appropriate, where AI risks are not managed.

What it requires

APRA’s findings are blunt: adoption is accelerating and governance is lagging. It found information security gaps (new attack paths such as prompt injection and data leakage, identity and access controls that do not account for AI agents), governance that treats AI as “just another technology” and misses its adaptive behaviour and bias, heavy dependence on single suppliers with opaque upstream dependencies, and assurance that is point-in-time rather than continuous.

Against that, the letter sets expectations at two levels. Boards must “maintain sufficient understanding and literacy with respect to AI in order to set strategic direction and provide effective challenge and oversight”, and must oversee an AI strategy consistent with the entity’s risk appetite, with monitoring, reporting and clearly defined triggers for action when AI is not operating as expected. Management must actively manage AI-specific security vulnerabilities, put in place “frameworks (policy, standard, guidance) and reporting lines to promote safe, responsible and sustainable adoption of AI” with clear ownership across the AI lifecycle and staff training, keep “visibility over the full AI supply chain, including material, third-party and fourth-party dependencies”, and run assurance using recognised control frameworks with continuous monitoring proportionate to how critical each use is. No deadlines are set; APRA says it is finalising its supervisory plan and invites early engagement where an entity has heightened concerns.

Does this reach your business?

Only if APRA regulates you: an authorised deposit-taking institution, a general, life or private health insurer, or a registrable superannuation entity. If you are one of those, treat the letter as the standard your next supervisory conversation about AI will be measured against. If you are not, it does not bind you, but it is the clearest statement any Australian regulator has made about what good AI governance looks like, and it maps closely onto what a customer in financial services will ask of its suppliers.

What we recommend

For a regulated entity, our advice is to start with the two things APRA can check quickly: a board that can show it understands the AI the business uses, and a named owner for every AI system across its lifecycle. Then work through the four gaps the letter names, in order: an inventory of AI systems and their suppliers (including who sits behind the supplier), controls for AI-specific attack paths, monitoring that runs continuously rather than at sign-off, and a fallback for any critical operation that AI supports. For a business outside APRA’s reach that sells to one, our advice is to expect these questions in due diligence and to keep the same record: what AI you use, who owns it, what it touches, and how you would know if it misbehaved.