Rapid Review into Australian Government arrangements for an AI‑driven cyber incident

Executive summary

This Review was triggered by the widely reported OpenAI incident affecting Services Australia. While focused on that specific event, it considers whether the Australian Government is prepared for risks driven by artificial intelligence (AI) more broadly. The Review is timely, providing an early real-world example of the opportunities, risks and governance challenges associated with advanced AI. These will be vital as the Australian Government seeks to maximise the benefits of AI, manage its risks, and strengthen Australia's influence over its safe development and deployment through initiatives including the AI Standards.

In Australia, like elsewhere, AI is at the early stages of transforming society, bringing significant economic, scientific and social benefits. Further profound changes are likely. At the same time, increasingly capable AI models are creating and amplifying additional risks at a faster pace, including risks to Government systems and processes. Advancing AI systems are likely to continue to exhibit behaviours that are difficult to predict, monitor and control. The full extent and scope of such risks are not yet known, including to the very companies developing the technology. Incidents of this nature were warnings months ago and are a reality now. More incidents are likely, as shown by subsequent revelations from OpenAI and Anthropic about comparable incidents outside Australia.

The Review identifies opportunities to strengthen Australia's preparedness and response to the challenges posed by advanced AI systems. These include measures to incentivise safe AI development and deployment, improve collaboration with AI firms, enhance preparedness for AI-related crises, and strengthen the Australian Government's cyber security posture in an evolving threat environment.

Australian Government efforts will rely on constructive cooperation underpinned by requirements for prompt and detailed information sharing by frontier AI labs. AI developers have the greatest visibility of their models’ capabilities, behaviours and emerging risks. This basic arrangement – of close cooperation with industry – is the foundation for our existing, world-leading approach to cyber security, and should be central to managing the risks and benefits of AI.

While OpenAI’s engagement with the Australian Government has been constructive since the incident was first disclosed to Services Australia, the Review finds the initial timeliness and method of disclosure were inadequate. The Australian Government expects and should demand major improvements in incident reporting from AI companies. Australia’s approach to AI must ensure collaboration and accountability work together. Given market incentives to develop increasingly capable AI systems quickly, regulation should create equally strong incentives for safety, backed by meaningful consequences for harmful conduct, and international cooperation to strengthen standards and oversight.

The OpenAI incident affecting Services Australia highlighted the challenge AI-related incidents pose for established government coordination and decision-making processes. While incident-management processes are well designed and tested for malicious cyber activity, the Review finds they are not designed to identify and respond to novel incidents of this kind. Uncertainty about whether and to what extent this incident was novel, meant its broader policy, regulatory and strategic implications were not considered early across Government. The Review concludes that information sharing between technical agencies, enabling agencies, and the National Cyber Security Coordinator should be strengthened to ensure coordination and information sharing with broader stakeholders can occur quickly, supporting decision makers to respond to incidents with significant public interest implications. More broadly, AI will alter the likelihood, severity, and consequences across hazard types (not just cyber security), so agencies with responsibility for leading the response to incidents need to reassess risk, coordination and response plans.

The Review also considered whether Australian Government systems were appropriately preparing for a substantially degraded cyber threat environment owing to the impacts of AI. The Review finds existing cyber security fundamentals remain essential but should be complemented by greater use of AI to strengthen cyber defence. Agencies should also assess how AI-enabled and autonomous threats may change targeting patterns and attack pathways. This reinforces the need to reduce legacy technology, limit internet-facing systems where possible, and continue the transition towards secure, centralised data environments in government. In other words, for government and for Australia more broadly, our new AI-influenced operating environment makes effective cyber defence even more imperative.

The significance of this incident lies less in its immediate cyber impacts than in what it revealed about Australia's preparedness for increasingly capable AI systems. It provides a valuable, galvanising opportunity to address gaps now, helping ensure Australia is better prepared for incidents of greater scale, complexity and consequence.