THE ESSENTIALS IN 30 SECONDS

  • APRA's System Risk Outlook, published on 12 November 2026, says frontier models from Anthropic and OpenAI are increasingly able to find and exploit software flaws.
  • The 2027 System Stress Test will model AI-driven disruption for banks, superannuation funds and life insurers, including a sharp fall in AI-related asset values.
  • APRA says it does not regulate AI itself but will step up supervision of how institutions use it and of shared dependencies on providers.

The Australian Prudential Regulation Authority (APRA), which supervises the country's banks, insurers and superannuation funds, published its twice-yearly System Risk Outlook on 12 November. This edition places artificial intelligence at the centre of its assessment, and it confirms that the regulator's next system-wide stress test, in 2027, will look at AI-driven shocks, which ABC News describes as a first for APRA.

What the stress test will cover

According to the report, the 2027 System Stress Test will involve banks, super funds and life insurers, as well as the wider financial system. It will examine how AI-driven economic disruption could affect them under severe but plausible scenarios. APRA names two: rapid AI adoption that undermines some business models, and a sharp downward revaluation of AI-related assets. Speaking at a media briefing the same day, APRA chair John Lonsdale said, as reported by ABC News, that large institutions were not yet as prepared as they should be, adding that it is "pretty safe to assume there will be interruptions".

Frontier models and cyber risk

The report says frontier models from Anthropic and OpenAI are becoming more capable of finding software flaws, writing exploits and automating complex cyber tasks. It notes that major US technology companies have reported a sharp rise in public vulnerability disclosures since April 2026, leaving institutions and their service providers less time to patch. It also cites a June 2026 case in which an OpenAI agent got around controls and reached an Australian government website.

APRA has been building towards this for months. On 30 April it called for a step change in AI risk management, naming Anthropic's Claude Mythos as a model expected to raise the speed and scale of cyber attacks. Over June and July, APRA and the corporate regulator ASIC held nine roundtables on frontier AI with more than 380 entities.

A technology-neutral approach

APRA stresses that it does not regulate AI technology directly. Instead, it looks at four channels: AI used inside institutions, AI used against them, shared dependencies on common models or providers, and wider effects on jobs, asset prices and business models. Concentration is a recurring concern: 80 of the 100 most-cited material service providers are based or mainly operate outside Australia. The report also flags that clustered data centre projects are stretching available insurance capacity. Details of the stress test scenarios have not yet been published.