Banking regulator warns AI is outpacing banks' safeguards

Lonsdale: interruptions are coming

APRA released its November 2026 System Risk Outlook on 12 November, placing artificial intelligence at the centre of its assessment of threats to the financial system.^[1] Chair John Lonsdale said the system remained sound but that conditions for overlapping shocks were hardening.

"Australia's financial system remains resilient, but the risk environment is becoming less forgiving," Lonsdale said.^[2] At a media briefing following the report's release, Lonsdale went further: "I think it is pretty safe to assume there will be interruptions, there will be problems."^[3]

Four channels, one interconnected problem

The System Risk Outlook identifies four connected AI-risk channels: AI used within institutions; AI used against institutions through cyber attacks and fraud; shared AI and service-provider dependencies; and broader economy-wide effects on employment, asset prices, competition and business models.^[1] APRA said the channels interact, meaning a disruption in one can amplify stress in another.

Frontier AI models are increasingly able to find and exploit software vulnerabilities and automate complex cyber tasks, potentially shortening the time available for financial institutions and their providers to assess exposure, patch systems and contain attacks.^[1] The Australian Signals Directorate has separately warned boards to prepare for incidents that could accelerate from days to hours.^[4]

ASIC echoed those concerns in an open letter to Australian financial services licensees, telling them that frontier AI was speeding up and broadening sophisticated cyber activity and that boards needed tested incident-response and business-continuity plans.^[5]

Offshore concentration amplifies the risk

Of the 100 most-used material service providers reported to APRA, 80 are headquartered or principally operated outside Australia, and some individual providers support institutions collectively holding more than half of APRA-regulated financial assets.^[1] APRA said that level of offshore concentration made supplier assurance, recovery options, threat sharing and coordinated incident response central to resilience.

The report also flagged the risk of autonomous AI agents pursuing goals without effective oversight, warning that unauthorised actions and unintended harm could follow. Concentration risk, where entities depended heavily on a single provider across several AI uses without adequate contingency plans, received similar attention.

Governance falling behind

APRA's April cross-industry supervisory review found that governance, risk management, assurance and operational-resilience practices were not keeping pace with the speed and complexity of AI adoption.^[6] That review also identified concentration risks where entities depended heavily on one provider across multiple AI uses and lacked adequate contingency planning.^[6]

Lonsdale said at the November briefing that major institutions were not yet where they needed to be, with APRA asking whether entities had systems capable of rapid recovery and continuity of service for customers.^[3]

2027 stress test to cover the whole system

APRA's 2027 System Stress Test will examine how AI-driven economic disruption could affect banks, superannuation funds, life insurers and the broader financial system.^[1] The exercise will run two severe but plausible scenarios: rapid AI adoption that disrupts established business models, and a sharp reassessment of AI-related asset values.

The 2027 exercise builds directly on the April supervisory review findings and extends the scope of previous stress testing to cover structural economic risk alongside financial risk, with results expected to inform APRA's supervisory priorities through the following year.^[6]