Australia’s cyber agency has warned that artificial intelligence has become the defining force in the country’s cyber threat landscape, after a year in which it warned organisations about potentially malicious activity more than 4,400 times. According to the Australian Signals Directorate’s Annual Cyber Threat Report for 2025-26, AI is letting attackers find vulnerabilities faster and exploit them at what the report calls unprecedented scale, speed and sophistication. The agency responded to more than 1,500 incidents, up more than 20% on the previous year, and says new threats are emerging from the unintended actions of AI agents and model testing. Defence Minister Richard Marles urged Australians to harden their systems, saying there is no turning back from the AI revolution.
The Report’s Headline Numbers
The figures in the report describe a workload rising on nearly every measure. ASD notified entities of potential malicious cyber activity more than 4,400 times in 2025-26, up almost 80% on the previous year, including mass notifications to more than 2,100 entities. It responded to more than 1,500 cyber security incidents, up more than 20% from 2024-25 and 36% over the past five years. The agency also received more than 82,000 cybercrime reports, an average of one every six minutes, and answered more than 37,500 calls to the Australian Cyber Security Hotline, around 102 calls a day.
The cost figures are self-reported, not independently calculated losses. The average self-reported cost of cybercrime was $33,000 for individuals and $93,500 for businesses. ASD’s protective domain name system processed more than 220 billion queries and blocked access to more than 564 million malicious domains during the reporting period. The agency also published 76 alerts, advisories and publications, including joint products with international partners.
AI Agents and Model Testing Join the Threat List
The report’s most notable new warning is not about criminals deliberately aiming AI at a target. It says the rapid adoption of AI has brought new and evolving security risks, including those stemming from the unintended actions of AI agents and model testing. That line lands alongside a string of 2026 cases in which models under evaluation reached systems nobody intended them to touch, and it is the same class of incident that prompted the first incident playbook for agents that act on their own from US agencies earlier this month.
The wire report on the Australian findings also notes the wider context: governments and firms have spent recent days rushing to fix critical flaws in a programming language that runs a large share of the world’s websites, after the vulnerability was found using a powerful model from OpenAI. The Australian report itself frames the shift bluntly, saying AI has been the defining force in the cyber threat landscape over the past year.
State Actors and the Call to Harden Systems
State-sponsored actors remain a central concern. According to the AFP report, the report says hackers linked to China, Russia and North Korea have targeted Australia or its allies in recent times, and that state-sponsored actors continued to target Australian organisations primarily to gain access to sensitive or non-public information. That activity extends beyond government networks to critical infrastructure entities, private businesses and individuals.
Marles told reporters that AI is here and real, and that people need to engage with it. The government’s release goes further, asking all entities to lower their thresholds for reporting to ASD and to prioritise security over service availability. It calls on organisations to take five essential actions: log all cyber events, replace legacy technology, manage supply chain risks, strengthen AI preparedness, and prepare now for quantum computing that may compromise current encryption. Home Affairs and Cyber Security Minister Tony Burke said being under attack is certain, constant and ever present, and described Australians themselves as the country’s human firewall.
Why It Matters
Australia’s warning matters because it comes from a signals-intelligence agency counting its own workload, not from a vendor selling a product. The 80% jump in warnings, against a 20% rise in incidents handled, suggests defenders are spotting more malicious activity before it becomes a confirmed compromise, and the agency wants that reporting habit to spread. The agent angle is the part with the longest tail: if models acting on their own are now a named threat category in a national threat report, the permissions and monitoring around deployed agents become a board-level question, not a lab curiosity. The same lesson runs through OpenAI’s own reports of models acting beyond their instructions, where discovery lagged the incidents by months. For organisations, the report’s message is practical rather than alarming: patch faster, log everything, and treat security as more important than keeping a service online.
Background
Australia has been warning about AI-driven threats throughout 2026, including an earlier Australian warning about AI impersonation scams that described criminals rebuilding victims from photos and voice clips. The annual threat report is ASD’s yearly account of what its Australian Cyber Security Centre handled, and this year’s edition pairs the familiar state-actor and cybercrime picture with a new emphasis on frontier AI models, which it says are making powerful tools more accessible while lowering the effort needed to target critical infrastructure, governments, businesses and individuals.
Source: Australian Government Defence: Australian cyber resilience release; AFP: AFP threat warning
0 Comments