Particle.news

ASD Says Frontier AI Has Reshaped Australia’s Cyber Threats

The agency warns advanced models let criminals and state-backed hackers find and exploit software flaws faster, raising the prospect of more outages and financial harm.

Overview

  • The Australian Signals Directorate published its Annual Cyber Threat Report on Monday showing a sharp rise in attacks, with more than 82,000 cybercrime reports, about 1,500 incidents and over 4,400 warnings to organisations.
  • Frontier AI models have been used to discover high‑risk software flaws, including a critical PHP vulnerability that forced widespread patching and temporary website takedowns.
  • The report names China, Russia and North Korea as routine targets of state‑sponsored hacking that can collect intelligence or preposition access that could be used to disrupt services.
  • ASD is using AI defensively, has set up early‑warning protocols with major model operators and Five Eyes partners, and carried out offensive actions such as disconnecting an Amadey controller and destroying millions of stolen files.
  • The agency urges organisations to remove exposed legacy systems, prioritise security over continuous availability, rotate credentials, extend logging and isolate test agents to reduce the growing risk to services and personal savings such as superannuation.