Preparing for Advanced AI Cyber Security Threats
Prepare now for a material increase in the speed, scale and sophistication of AI-enabled cyber attacks — on a timeframe of months, not years.
At a glance
- ASD and Five Eyes warn that advanced frontier and agentic AI will materially raise the speed, scale and sophistication of attacks — months, not years.
- The economics shift: probing thousands of systems approaches the cost of probing one. Assume every exposed system is continuously tested by automated adversaries.
What has changed
Generative AI already accelerated phishing, reconnaissance, vulnerability research and malicious-code development. Frontier and agentic systems go further — increasingly able to:
- identify vulnerabilities and reason across the stages of an attack;
- generate and modify attack techniques, and conduct reconnaissance at machine speed;
- act autonomously, chain actions toward a goal, and adapt to previous results.
ASD notes evidence of models completing complex, multi-step simulated intrusions. Work that once required substantial human expertise and time is becoming automated, scalable and inexpensive.
Recent Australian Government guidance
Frontier AI
Capable frontier models will disrupt the cyber-security environment; strengthen fundamental controls and Secure-by-Design.
cyber.gov.auCareful adoption
Agents combine data access, memory, tools, credentials and autonomy — adding excessive-privilege, prompt-injection, unintended-action, disclosure and cascading-failure risks.
cyber.gov.auMonths, not years
Reduce attack surface, strengthen identity, remove legacy technology, accelerate remediation and improve incident response.
cyber.gov.auA change in the threat model
Detecting individual attackers.
Systems that stay secure under continuous automated examination.
Assume that every externally accessible system can be continuously tested by intelligent automated adversaries.