Australia’s signals intelligence and cybersecurity agency, the Australian Signals Directorate (ASD), has published new guidance for boards of directors and senior leaders on managing frontier AI cyber risks.
The guidance warns that organizations failing to respond to frontier AI threats leave themselves exposed to current and emerging cyber risks, urging boards to press management to act now, including by supporting targeted investments to strengthen cybersecurity and resilience.
Frontier AI models have been widely available since early 2023 and continue to rapidly evolve. According to the guidance, frontier AI has the potential to reduce vulnerability discovery and exploitation timelines from days to hours while lowering the barriers for malicious actors.
The guidance recommends prioritizing cybersecurity strategies across immediate, short, medium and longer-term horizons.
Immediate priorities include securing attack surfaces and reducing software vulnerabilities. Short-term priorities focus on replacing legacy systems, reinforcing identity, credential and access management, restricting unnecessary privileges, and preparing for cyber security incidents. The medium-term priority is adopting AI for cyber defence, while the longer-term priority is modernizing systems using Secure by Design and Secure by Default principles.
The agency calls on organizations to review how advances in frontier AI models could affect their cyber security posture, assess the risk of relying on AI providers, including concerns over foreign ownership, control and influence and cyber supply chain risks, prepare for cyber incidents in an agentic AI-driven threat environment and strengthen governance, processes and capabilities for managing AI use.
“Frontier AI models are currently, and will continue to, fundamentally transform both offensive and defensive cyber capabilities,” the guidance said.
The guidance also provides recommendations to help organizations prevent, detect, respond to and recover from AI-related cybersecurity incidents, explaining how frontier AI models are making cyberattacks faster, more effective and easier to carry out.
The guidance further notes that frontier AI models can rapidly identify and weaponize vulnerabilities, combine multiple low-severity weaknesses into high-impact compromises, and perform malicious cyber activities with little or no human oversight.
Also Read: White House AI Framework Exempts Open-Weight Models from Pre-Release Review
