On this page
AI is reshaping the cyber threat environment. While it can strengthen cyber defence, it also enables vulnerabilities to be identified and exploited faster, with less predictability, at greater scale, and with increasing sophistication. As AI capabilities advance, cyber risks are likely to increase accordingly.
Although the consequences were limited, the OpenAI incident affecting Services Australia is a real-world manifestation of a previously recognised AI safety risk, and also highlights a changing risk exposure. The ability of capable AI systems to identify and exploit vulnerabilities is altering traditional assumptions about cyber threats, threat actors, and potential risk targets. AI agents may autonomously attack targets that would not typically be identified as high-value by human threat actors. As a result, the pool of potential targets may expand significantly, with capable AI models able to identify and exploit vulnerabilities across a much broader range of targets, at scale, and with little warning.
To respond to this shift away from traditional cyber security risk-assessment methodology, and as AI capabilities continue to develop, the Australian Government will need to ensure its cyber security posture evolves accordingly. This will require resilient systems, strong technical controls, effective vulnerability management, the use of AI proactively for defence, and continued adaptation to AI-enabled and increasingly autonomous threats.
There were several factors that meant the Services Australia Medicare Statistics Reporting Service was vulnerable to this unauthorised activity (or would be vulnerable to cyber attack).
- It was a public facing, unauthenticated internet-accessible website which expanded its attack surface
- It was a legacy system that only required minimal ongoing support, because it contained low-risk information and was therefore not prioritised for cyber security uplift.11
In June 2026, Five Eyes Cyber Security Agencies released a statement which urgently called for prioritising cyber resilience actions, including reducing attack surfaces, accelerating patching processes, addressing legacy systems, reviewing and strengthening identity and access controls, and preparing for incidents.xx As AI enables increased speed and scale in identifying and exploiting vulnerabilities, it raises the urgency for best practice to be implemented across government.
The Review reiterates that agencies must focus on the fundamental elements of good cyber security practices. These include the Essential Eight (particularly reaching Maturity Level 2),12 and engagement with ASD as it rolls out the revised approach through the new Essentials series. In particular, this requires authentication and application controls, and prioritising rapid patching cycles to meet demands of changing timeframes between vulnerability discovery and exploitation. The 2025 Commonwealth Cyber Security Posture report found that 22 per cent of Commonwealth agencies had reached Maturity Level 2.xxi While this is an increase from 15 per cent in 2024, most agencies are not reaching the cyber security standards required under the PSPF.xxii
In addition, the incident highlights the importance of shifting away from legacy technology as it is more vulnerable, owing to limited vendor support and patches. Legacy technology also has higher maintenance costs than newer systems, placing additional pressures on stretched departmental budgets. This is supported by a series of PSPF directions and requirements – including PSPF Direction 002-2026 – Strengthening Commonwealth Cyber Posture Against AI-Enabled Risks, undertaking a Technology Asset stocktake, implementing a technology security risk management plan for internet-facing systems, and implementing temporary mitigations for legacy systems while they are being phased out. Reducing the size of the Australian Government’s internet-facing attack surface will also be necessary to not only simplify the digital estate, but to mitigate against the new risks that agentic AI may pose to these previously low-value targets.
Recommendations
- Commonwealth agencies renew their focus on the fundamental elements of good cyber security practice, including implementation of ASD’s Essentials.
- Commonwealth agencies work to reduce the size of the government digital estate, consolidating digital systems, and shutting down or segregating systems that do not need to be connected to the internet.
- The Australian Government consider how to further harden cyber defence of Systems of Government Significance, including considering a changing threat environment and risk profile.
At the broadest level, Horizon 1 of the Cyber Security Strategy 2023-2030 has already delivered meaningful cyber security uplift across the government.xxii This work is continuing and deepening under Horizon 2, which aims to shift the government’s cyber security response to a proactive, investment-led approach.xxiv This includes through prioritised investment for Systems of Government Significance – the Commonwealth’s most critical digital services.
In addition to cyber security fundamentals, ASD has encouraged entities to use AI to strengthen defence.xxv To scale these efforts across government, ASD is developing guidance and assistance for other departments to run their own AI-powered code reviews.
In this respect, the Australian Government has been active in addressing AI-enabled cyber security risk, including through targetted risk-based cyber security uplift, which will help to protect against increasing AI cyber capabilities. The Government has also published specific guidance in response to frontier AI cyber developments. ASD has published a significant number of reports related to AI over several years, starting in 2023. More recently, ASD issued guidance on the impact of frontier AI on cyber security on 8 April 2026, the day after the release of Claude Mythos Preview on 7 April 2026. In addition, PSPF Advisory 001-2026 and PSPF Direction 002-2026 position cyber security in the context of advancing frontier AI cyber capabilities as a national priority, and provide practical advice from the Information Security Manual and ASD on hardening Commonwealth systems against frontier AI cyber threats.
While the OpenAI incident affecting Services Australia was unacceptable, it also highlights the strengths of the Government’s approach to cyber security and provided valuable lessons – both for government and OpenAI. While not the subject of this Review, it is noted that the Australian Institute of Health and Welfare was also subject to attempts by OpenAI models to gain unauthorised access to their systems, but was not compromised because of its strong cyber fundamentals. Agencies should consider lessons from this success. No government system should ever be compromised by third parties, whether intentionally or not, but the Medicare Statistics Reporting Service was a public, unauthenticated reporting service containing aggregate, low-sensitivity information, operating as a stand alone service on dedicated servers. The Medicare Statistics Reporting Service had controls that prevented access to broader datasets, which were commensurate to the threat landscape at the time of deployment. At the time of the OpenAI incident affecting Services Australia, the Medicare Statistics data was already flagged to be migrated to a more secure portal at data.gov.au. This transfer was accelerated in response to this incident.
Services Australia will continue to engage in substantial cyber security work, in line with the Government’s risk-based approach, noting its systems include critical digital services and high-value data holdings. Government has invested around $160 million for cyber security uplift for Services Australia. This is in addition to work underway to update Services Australia’s ICT architecture, which will further modernise and protect key ICT systems.
Action taken
- 2023 onward – ASD commenced publishing public advice on AI, and has regularly updated the Information Security Manual to include AI controls.
- May 2026 – Home Affairs issued guidance on cyber security readiness in the frontier AI era.
- June 2026 – ASD’s ACSC and Five Eyes released a joint statement warning that frontier AI models would fundamentally transform offensive and defensive cyber capabilities and recommended actions for organisations to address this risk.
- September 2026 – Home Affairs issued PSPF direction to require entities to undertake legacy technology stocktake, risk management plan, and incorporate new risk reduction measures for Systems of Government Significance (in response to AI’s impacts on the threat environment).
Recommendation
- Commonwealth agencies engage with ASD to identify and assess opportunities to responsibly use AI-enabled cyber defence tools and techniques to enhance cyber security monitoring, penetration testing, threat detection and incident response capabilities.
Footnotes
- This prioritisation was according to the low risk information contained.Return to footnote 11 ↩
- It is a requirement under the PSPF (Requirement 0105) that Commonwealth entities reach Maturity level 2 under the Essential Eight – a framework developed by the ASD to help organisations protect against cyber threats. ASD has recently updated this framework to a new Essentials series that builds upon the original Essential Eight to address contemporary cyber threats and modern enterprise environments. ‘Maturity levels’ are now transitioning into ‘priority phases’ under the Essentials. Organisations that achieved Maturity Level Two can expect many existing controls to contribute to actions in the early priority phases.Return to footnote 12 ↩